DO
OU 160413 >m
&SMANIA UNIVERSITY LIBRARY
53. O  Accession No. / 6 S 9
^hould be returned on or beforthe date last marked below
ASKOMPANEYETS
TH
RETICAL
PHYSICS
FOREIGN LANGUAGES PUBLISHING HOUSE
Moscow 1961
TRANSLATED FROM THE RUSSIAN
EDITED BY GEORGE YANKOVSKY
This translation has been read and approved
by the author, Professor A. S. Kompaneyets
Printed in the Union of Soviet Socialist Republics
CONTENTS
Page
From the Preface to the First Edition 7
Preface to the Second Edition 9
Part I. Mechanics 11
Sec. 1. Generalized Coordinates 11
See. 2. Lagrange's Equation 13
Sec. 3. Examples of Lagrange's Equations 24
Sec. 4. Conservation Laws 30
Sec. 5. Motion in a Central Field 41
Sec. 6. Collision of Particles 48
Sec. 7. Small Oscillations 57
Sec. 8. Rotating Coordinate Systems. Tnertial Forces 66
Sec. 9. The Dynamics of a Rigid Body 73
Sec. 10. General Principles of Mechanics 81
Part II. Electrodynamics 92
Sec. 11. Vector Analysis 92
Sec. 12. The Electromagnetic Field. Maxwell's Equations 104
Sec. 13. The Action Principle for the Electromagnetic Field 117
Sec. 14. The Electrostatics of Point Charges. Slowly Varying Fields . . 124
Sec. 15. The Magnetostatics of Point Charges 135
Sec. 16. Electrodynamics of Material Media 144
Sec. 17; Plane Electromagnetic Waves 162
Sec. 18. Transmission of Signals. Almost Plane Waves 173
Sec. 19. The Emission of Electromagnetic Waves 181
Sec. 20. The Theory of Relativity 190
Sec. 21. Relativistic Dynamics 211
Part III. Quantum Mechanics 229
Sec. 22. The Inadequacy of Classical Mechanics.
The Analogy Between Mechanics and Geometrical Optics 229
Sec. 23. Electron Diffraction 238
Sec. 24. The Wave Equation 244
6 CONTENTS
Page
Sec. 25. Certain Problems of Quantum Mechanics 252
Sec. 26. Harmonic Oscillatory Motion in Quantum Mechanics
(Linear Harmonic Oscillator) 265
Sec. 27. Quantization of the Electromagnetic Field 271
Sec. 28. QuasiClassical Approximation 280
Sec. 29. Operators in Quantum Mechanics 291
Sec. 30. Expansions into Wave Functions 301
Sec. 31. Motion in a Central Field 312
Sec. 32. Electron Spin 323
Sec. 33. ManyElectron Systems 334
Sec. 34. The Quantum Theory of Radiation 353
Sec. 35. The Atom in a Constant External Field 368
Sec. 36. Quantum Theory of Dispersion 379
Sec. 37. Quantum Theory of Scattering 385
Sec. 38. The Relativistic Wave Equation for an Electron 394
Part IV. Statistical Physics 413
Sec. 39. The Equilibrium Distribution of Molecules in an Ideal Gas . . 413
Sec. 40. Boltzmann Statistics (Translational Motion of a Molecule. Gas
in an External Field) 430
Sec. 41. Boltzmann Statistics (Vibrational and Rotational Molecular
Motion) 447
Sec. 42. The Application of Statistics to the Electromagnetic Field and
to Crystalline Bodies 457
Sec. 43. Bose Distribution 474
Sec. 44. Fermi Distribution 477
Sec. 45. Gibbs Statistics 498
Sec. 46. Thermodynamic Quantities 512
Sec. 47. The Thermodynamic Properties of Ideal Gases in Boltzmann
Statistics 535
Sec. 48. Fluctuations 546
Sec. 49. Phase Equilibrium 557
Sec. 50. Weak Solutions 568
Sec. 51. Chemical Equilibria 576
Sec. 52. Surface Phenomena 582
Appendix . 586
Bibliography 588
Subject Index 589
FROM THE PREFACE TO THE FIRST EDITION
This book is intended for readers who are acquainted with the
course of general physics and analysis of nonspecializing institutions
of higher education. It is meant chiefly for engineer physicists, though
it may also be useful to specialists working in fields associated with
physics chemists, physical chemists, biophysicists, geophysicists,
and astronomers.
Like the natural sciences in general, physics is based primarily
on experiment, and, what is more, on quantitative experiment.
However, no series of experiments can constitute a theory until a
rigorous logical relationship is established between them. Theory not
only allows us to systematize the available experimental material, but
also makes it possible to predict new facts which can be experimentally
verified.
All physical laws are expressed in the form of quantitative relation
ships. In order to interrelate quantitative laws, theoretical physics
appeals to mathematics. The methods of theoretical physics, which
are based on mathematics, can be fully mastered only by those who
have acquired a very considerable volume of mathematical knowledge.
Nevertheless, the basic ideas and results of theoretical physics are
readily comprehensible to any reader who has an understanding of
differential and integral calculus, and is acquainted with vector
algebra. This is the minimum of mathematical knowledge required
for an understanding of the text that follow r s.
At the same time, the aim of this book is not only to give the reader
an idea about what theoretical physics is, but also to furnish him
with a working knowledge of the basic methods of theoretical physics.
For this reason it has been necessary to adhere, as far as possible,
to a rigorous exposition. The reader will more readily agree with
the conclusions reached if their inevitability has been made obvious
to him. In order to activize the work of the student, some of the
applications of the theory have been shifted into the exercises, in
which the line of reasoning is not so detailed as in the basic text.
In compiling such a relatively small book as this one it has been
necessary to cut down on the space devoted to certain important
8 FROM THE PREFACE TO THE FIRST EDITION
sections of theoretical physics, and omit other branches entirely.
For instance, the mechanics of solid media is not included at all
since to set out this branch, even in the same detail as the rest of
the text, would mean doubling the size of the book. A few results
from the mechanics of continuous media are included in the exercises
as illustrations in thermodynamics. At the same time, the mechanics
and electrodynamics of solid media are less related to the fundamental,
gnosiological problems of physics than microscopic electrodynamics,
quantum theory, and statistical physics. For this reason, very little
space is devoted to macroscopic electrodynamics: the material has
been selected in such a way as to show the reader how the transition
is made from microscopic electrodynamics to the theory of quasi
stationary fields and the laws of the propagation of light in media.
It is assumed that the reader is familiar with these problems from
courses of physics and electricity.
On the whole, the book is mainly intended for the reader who is
interested in the physics of elementary processes. These considerations
have also dictated the choice of material ; as in all nonencyclopaedic
manuals, this choice is inevitably somewhat subjective.
In compiling this book, I have made considerable use of the excellent
course of theoretical physics of L. D. Landau and E. M. Lifshits. This
comprehensive course can be recommended to all those who wish
to obtain a profound understanding of theoretical physics.
I should like to express my deep gratitude to my friends who have
made important observations: Ya. B. Zeldovich, V. G. Levich,
E. L. Feinberg, V. I. Kogan and V. I. Goldansky.
A. Kompaneyets
PREFACE TO THE SECOND EDITION
In this second edition I have attempted to make the presentation
more systematic and rigorous without adding any difficulties. In
order to do this it has been especially necessary to revise Part III,
to which I have added a special section (Sec. 30) setting out the general
principles of quantum mechanics; radiation is now considered only
with the aid of the quantum theory of the electromagnetic field,
since the results obtained from the correspondence principle do not
appear sufficiently justified.
Gibbs' statistics are included in this edition, which has made it
necessary to divide Part IV into something in the nature of two
cycles: Sec. 3944, where only the results of combinatorial analysis
are set out, and Sec. 4552, an introduction to the Gibbs' method,
which is used as background material for a discussion of thermo
dynamics. A phenomenological approach to thermodynamics would
nowadays appear an anachronism in a course of theoretical physics.
In order not to increase the size of the book overmuch, it has been
necessary to omit the theory of beta decay, the variational properties
of eigenvalues, and certain other problems included in the first
edition.
I am greatly indebted to A. F. Nikiforov and V. B. Uvarov for
pointing out several inaccuracies in the first edition of the book.
A. Kompaneyets
PART I
MECHANICS
See. 1. Generalized Coordinates
Frames of reference. In order to describe the motion of a mechanical
system, it is necessary to specify its position in space as a function
of time. Obviously, it is only meaningful to speak of the relative
position of any point. For instance, the position of a flying aircraft
is given relative to some coordinate system fixed with respect to the
earth; the motion of a charged particle in an accelerator is given
relative to the accelerator, etc. The system, relative to which the
motion is described, is called a frame of reference.
Specification of time. As will be shown later (Sec. 20), specification
of time in the general case is also connected with defining the frame
of reference in which it is given. The intuitive conception of a uni
versal, unique time, to which we are accustomed in everyday life,
is, to a certain extent, an approximation that is only true when the
relative speeds of all material particles are small in comparison
with the velocity of light. The mechanics of such slow movements
is termed Newtonian, since Isaac Newton was the first to formulate
its laws.
Newton's laws permit a determination of the position of a mechanical
system at an arbitrary instant of time, if the positions and velocities
of all points of the system are known at some initial instant, and also
if the forces acting in the system are known.
Degrees of freedom of a mechanical system. The number of inde\
pendent parameters defining the position of a mechanical system in
space is termed the number of its degrees of freedom.
The position of a particle in space relative to other bodies is defined
with the aid of three independent parameters, for example, its
Cartesian coordinates. The position of a system consisting of N
particles is determined, in general, by 3JV independent parameters.
However, if the distribution of points is fixed in any way, then
the number of degrees of freedom may be less than 3^. For example,
12 MECHANICS [Part I
if two points are constrained by some form of rigid nondeformable
coupling, then, upon the six Cartesian coordinates of these points,
x i> Vi> z i> x z> y%> Z 2> * s imposed the condition
(x 2 x l ^ + (y 2 ~y l ^ + (z 2 z 1 )^Rl 29 (1.1)
where J? 12 is the given distance between the points. It follows that
the Cartesian coordinates are no longer independent parameters:
a relationship exists between them. Only five of the six values
x l9 . . . , Z 2 are now independent. In other words, a system of two
particles, separated by a fixed distance, has five degrees of freedom. If
we consider three particles which are rigidly fixed in a triangle, then
the coordinates of the third particle must satisfy the two equations :
y^ + (z,z 1 )^Rl l , (1.2)
y 2 )* + (Zsz 2 )*=:Rl,. (1.3)
Thus, the nine coordinates of the vertices of the rigid triangle are
defined by the three equations (1.1), (1.2) and (1.3), and hence only
six of the nine quantities are independent. The triangle has six
degrees of freedom.
The position of a rigid body in space is defined by three points
which do not lie on the same straight line. These three points, as we
have just seen, have six degrees of freedom. It follows that any rigid
body has six degrees of freedom. It should be noted that only such
motions of the rigid body are considered as, for example, the rotation
of a top, where no noticeable deformation occurs that can affect
its motion.
Generalized coordinates. It is not always convenient to describe
the position of a system in Cartesian coordinates. As we have already
seen, when rigid constraints exist, Cartesian coordinates must satisfy
supplementary equations. In addition, the choice of coordinate system
is arbitrary and should be determined primarily on the basis of
expediency. For instance, if the forces depend only on the distances
between particles, it is reasonable to introduce these distances into
dynamical equations explicitly and not by means of Cartesian
coordinates.
In other words, a mechanical system can be described by coordinates
whose number is equal to the number of degrees of freedom of the
system. These coordinates may sometimes coincide with the Cartesian
coordinates of some of the particles. For example, in a system of two
rigidly connected points, these coordinates can be chosen in the
following way: the position of one of the points is given in Cartesian
coordinates, after which the other point will always be situated on
a sphere whose centre is the first point. The position of the second
point on the sphere may be given by its longitude and latitude.
Sec. 2] LAGRANGE'S EQUATION 13
Together with the three Cartesian coordinates of the first point,
the latitude and longitude of the second point completely define
the position of such a system in space.
For three rigidly bound points, it is necessary, in accordance with
the method just described, to specify the position of one side of the
triangle and the angle of rotation of the third vertex about that
side.
The independent parameters which define the position of a
mechanical system in space are called its generalized coordinates.
We will represent them by the symbols g^, where the subscript a
signifies the number of the degree of freedom.
As in the case of Cartesian coordinates, the choice of generalized
coordinates is to a considerable extent arbitrary. It must be chosen
so that the dynamical laws of motion of the system can be formulated
as conveniently as possible.
Sec. 2. Lagrange's Equation
In this section, equations of motion will be obtained in terms of
arbitrary generalized coordinates. In such form they are especially
convenient in theoretical physics.
Newton's Second Law. Motion in mechanics consists in changes in
the mutual configuration of bodies in time. In other words, it is
described in terms of the mutual distances, or lengths, and intervals
of time. As was shown in the preceding section, all motion is relative;
it can be specified only in relation to some definite frame of refer
ence.
In accordance with the level of knowledge of his time, Newton
regarded the concepts of length and time interval as absolute, which
is to say that these quantities are the same in all frames of reference.
As will be shown later, Newton's assumption was an approximation
(see Sec. 20). It holds when the relative speeds of all the particles
are small compared with the velocity of light; here Newtonian
mechanics is based on a vast quantity of experimental facts.
In formulating the laws of motion a very convenient concept is
the material particle, that is, a body whose position is completely
defined by three Cartesian coordinates. Strictly speaking, this
idealization is not applicable to any body. Nevertheless, it is in
every way reasonable when the motion of a body is sufficiently well
defined by the displacement in space of any of its particles (for
example, the centre of gravity of the body) and is independent of
rotations or deformations of the body.
If we start with the concept of a particle as the fundamental
entity of mechanics, then the law of motion (Newton's Second Law)
is formulated thus:
14 MECHANICS [Part I
~
(2.1)
Here, F is the resultant of all the forces applied to the particle
i2 f
(the vector sum of the forces) r^ is the vector acceleration, the
Cartesian components of which are
d 2 z
The quantity m involved in equation (2.1) characterizes the particle
and is called its mass.
Force and mass. Equality (2.1) is the definition of force. However,
it should not be regarded as a simple identity or designation, be
cause (2.1) establishes the form of the interaction between bodies
in mechanics and thereby actually describes a certain law of nature.
The interaction is expressed in the form of a differential equation
that includes only the second derivatives of the coordinates with
respect to time (and not derivatives, say, of the fourth order).
In addition, certain limiting assumptions are usually made in
relation to the force. In Newtonian mechanics it is assumed that
forces depend only on the mutual arrangement of the bodies at the
instant to which the equality refers and do not depend on the con
figuration of the bodies at previous times. As we shall see later (see
Part II), this supposition about the character of interaction forces
is valid only when the speeds of* the bodies are small compared with
the velocity of light.
The quantity m in equality (2.1) is a characteristic of the body,
its mass. Mass may be determined by comparing the accelerations
which the same force imparts to different bodies; the greater the
acceleration, the less the mass. In order to measure mass, some body
must be regarded as a standard. The choice of a standard body is
completely independent of the choice of standards of length and
time. This is what makes the dimension (or unit of measurement)
of mass a special dimension, not related to the dimensions of length
and time.
The properties of mass arc established experimentally. Firstly, it
can be shown that the mass of two equal quantities of the same
substance is equal to twice the mass of each quantity. For example,
one can take two identical scale weights and note that a stretched '
spring gives them equal accelerations. If we join two such weights
and subject them to the action of the same spring, which has been
stretched by the same amount as for each weight separately, the
acceleration will be found to be one half what it was. It follows that
the overall mass of the weights is twice as great, since the force
depends only on the tension of the spring and could not have changed.
Sec. 2] LAGRANGE'S EQUATION 15
Thus, mass is an additive quantity, that is, one in which the whole
is equal to the sum of the quantities of each part taken separately.
Experiment shows that the principle of additivity of mass also
applies to bodies consisting of different substances.
In addition, in Newtonian mechanics, the mass of a body is a
constant quantity which does not change with motion.
It must not be forgotten that the additivity and constancy of
masses are properties that follow only from experimental facts which
relate to very specific forms of motion. For example, a very important
law, that of the conservation of mass in chemical transformations
involving rearrangement of the molecules and atoms of a body,
was established by M. V. Lomonosov experimentally.
Like all laws deduced from experiment, the principle of additivity
of mass has a definite degree of precision. For such strong interactions
as take place in the atomic nucleus, the breakdown of the additivity
of mass is apparent (for more detail see Sec. 21).
We may note that if instead of subjecting a body to the force of
a stretched spring it were subjected to the action of gravity, then
the acceleration of a body of double mass would be equal to the
acceleration of each body separately. From this we conclude that
the force of gravity is itself proportional to the mass of a body.
Hence, in a vacuum, in the absence of air resistance, all bodies fall
with the same acceleration.
Inertial frames of reference. In equation (2.1) we have to do with
the acceleration of a particle. There is no sense in talking about
acceleration without stating to which frame of reference it is referred.
For this reason there arises a difficulty in stating the cause of the
acceleration. This cause may be either interaction between bodies
or it may be due to some distinctive properties of the reference frame
itself. For example, the jolt which a passenger experiences when a
carriage suddenly stops is evidence that the carriage is in nonuniform
motion relative to the earth.
Let us consider a set of bodies not affected by any other bodies,
that is, one that is sufficiently far away from them. We can suppose
that a frame of reference exists such that all accelerations of the
set of bodies considered arise only as a result of the interaction between
the bodies. This can be verified if the forces satisfy Newton's Third
Law, i.e., if they are equal and opposite in sign for any pair of particles
(it is assumed that the forces occur instantaneously, and this is true
only when the speeds of the particles are small compared with the
speed of transmission of the interaction).
A frame of reference for which the acceleration of a certain set of
particles depends only on the interaction between these particles
is called an inertial frame (or inertial coordinate system). A free
particle, not subject to the action of any other body, moves, relative
to such a reference frame, uniformly in a straight line or, in everyday
16 MECHANICS [Part I
language, by its own momentum. If in a given frame of reference
Newton's Third Law is not satisfied we can conclude that this is
not an inertial system.
Thus, a stone thrown directly downwards from a tall tower is
deflected towards the east from the direction of the force of gravity.
This direction can be independently established with the aid of a
suspended weight. It follows that the stone has a component of
acceleration which is not caused by the force of the earth's attraction.
From this we conclude that the frame of reference fixed in the earth
is noninertial. The noninertiality is, in this case, due to the diurnal
rotation of the earth.
On the forces ot friction. In everyday life we constantly observe
the action of forces that arise from direct contacts between bodies.
The sliding and rolling of rigid bodies give rise to forces of friction.
The action of these forces causes a transition of the macroscopic
motion of the body as a whole into the microscopic motion of the
constituent atoms and molecules. This is perceived as the generation
of heat. Actually, when a body slides an extraordinarily complex
process of interaction occurs between the atoms in the surface layer.
A description of this interaction in the simple terms of frictional
forces is a very convenient idealization for the mechanics of macro
scopic motion, but, naturally, does not give us a full picture of the
process. The concept of frictional force arises as a result of a certain
averaging of all the elementary interactions which occur between
bodies in contact.
In this part, which is concerned only with elementary laws, we shall
not consider averaged interactions where motion is transferred to
the internal, microscopic, degrees of freedom of atoms and molecules.
Here, we will study only those interactions which can be completely
expressed with the aid of elementary laws of mechanics and which
do not require an appeal to any statistical concepts connected with
internal, thermal, motion.
Ideal rigid constraints. Bodies in contact also give rise to forces
of interaction which can be reduced to the kinematic properties of
rigid constraints. If rigid constraints act in a system they force
the particles to move on definite surfaces. Thus, in Sec. 1 we con
sidered the motion of a single particle on a sphere, at the centre of
which was another particle.
This kind of interaction between particles does not cause a transition
of the motion to the internal, microscopic, degrees of freedom of
bodies. In other words, motion which is limited by rigid constraints
is completely described by its own macroscopic generalized co
ordinates q a .
If the limitations imposed by the constraints distort the motion,
they thereby cause accelerations (curvilinear motion is always
accelerated motion since velocity is a vector quantity). This ac
Sec. 2] LAGRANGE'S EQUATION 17
celeration can be formally attributed to forces which are called
reaction forces of rigid constraints.
Reaction forces change only the direction of velocity of a particle
but not its magnitude. If they were to alter the magnitude of the
velocity, this would produce a change also in the kinetic energy of
the particle. According to the law of conservation of energy, heat
would then be generated. But this was excluded from consideration
from the very start.
To summarize, the reaction forces of ideally rigid constraints do
not change the kinetic energy of a system. In other words, they do
not perform any work on it, since work performed on a system is
equivalent to changing its kinetic energy (if heat is not gener
ated).
In order that a force should not perform work, it must be perpen
dicular to the displacement. For this reason the reaction forces of
constraints are perpendicular to the direction of particle velocity
at each given instant of time.
However, in problems of mechanics, the reaction forces are not
initially given, as are the functions of particle position. They are
determined by integrating equations (2.1), with account taken of
constraint conditions. Therefore, it is best to formulate the equations
of mechanics so as to exclude constraint reactions entirely. It turns
out that if we go over to generalized coordinates, the number of
which is equal to the number of degrees of freedom of the system,
then the constraint reactions disappear from the equations. In this
section we shall make such a transition and will obtain the equations
of mechanics in terms of the generalized coordinates of the system.
The transformation from rectangular to generalized coordinates.
We take a system with a total of 3N==n Cartesian coordinates of
which v are independent. We will always denote Cartesian coordinates
by the same letter xi, understanding by this symbol all the co
ordinates x, y, z ; this means that i varies from 1 to 3jY, that is, from 1
to n. The generalized coordinates we denote by # a (l^Ca^v). Since
the generalized coordinates completely specify the position of their
s\'stem, Xi are their unique functions:
x l ^x i (q l , ^ 2 , ... ? a , ...,v). (2.2)
From this it is easy to obtain an expression for the Cartesian com
ponents of velocity. Differentiating the function of many variables
Xi ( . . . g) with respect to time, we have
dt
In the subsequent derivation we shall often have to perform
summations with respect to all the generalized coordinates q at
20060
18 MECHANICS [Part I
and double and triple sums will be encountered. In order to save
space we introduce the following summation convention.
If a Greek symbol is met twice on one side of an equation, it will be
understood as denoting a summation from unity to v, that is, over all
the generalized coordinates. (It is not convenient to use this convention
for the Latin characters which denote the Cartesian coordinates.)
Then the velocity . can be rewritten thus:
_^ a '_ __ djri ^? /o Q\
dt ~~ #7a dt ' ' *~" '
Here the summation sign is omitted.
The total derivative with respect to time is usually denoted by a
dot over the corresponding variable:
d * __ . dq* .
~dt *" "dT^V" ( ~' '
In this notation, (2.3) is written in an even more abbreviated form:
f) .?** . i c\ f\
=t~q*. (2.5)
Differentiating (2.5) with respect to time again, we obtain an
expression for the Cartesian components of acceleration:
The total derivative in the first term is written as usual:
A
dt
The Greek symbol over which the summation is performed is denoted
by the letter (3 to avoid confusion with the symbol a, which denotes
the summation in the expression for velocity (2.5). Thus, we obtain
the desired expression for , :
8 2 xi . . , dxi .. /rt .
Xi =  q * q * + q '" (2 ' 6)
The first term on the righthand side contains a double summation
with respect to a and p.
Potential of a force. We now consider components of force. In
many cases, the three components of the vector of a force acting
on a particle can be expressed in terms of one scalar function U
according to the formula:
Sec. 2j LAGRANGE'S EQUATION 19
Such a function can always be chosen for the force of Newtonian
attraction, and for electrostatic and elastic forces. The function U
is called the potential of the force.
It is clear that by far not every system of forces can, in the general
case, be represented by a set of partial derivatives (2.7), since if
then we must have the equality
< F\ = _ ?F k _ <*V
c^k Pxi c'.r, cVfc
for all i, k, which is not, in advance, obvious for the arbitrary functions
F, , Fk . The definite form of the potential, in those cases where it
exists, will be given below for various forces.
Expression (.7) defines the potential function U to the accuracy
of an arbitrary constant term. U is also called the potential energy
of the system. For example, the gravitational force F = mg,
while the potential energy of an elevated body is equal to mgz,
where {7^980 cm/sec 2 is the acceleration of a freely falling body
and z is the height to which it has been raised. It can be calculated
from any level, which in the given case corresponds to a determi
nation of U to the accuracy of a constant term. A more precise ex
pression for the force of gravity than F mg (with allowance
made for its dependence on height also admits of a potential, which
we shall derive a little later [see (3.4)].
We denote the component reaction forces of rigid constraints by
Fi. We now note that
n
f"*dxi=0 9 (2.8)
i i
if the displacements are compatible with the constraints. Indeed,
(2.8) expresses precisely the work performed by the reaction forces
for a certain possible displacement of the system; but this work
has been shown to be equal to zero.
Lagrange's equations.* We will now write down the equations of
motion with the aid of (2.7) and (2.8) as
fcfc +) J^ <"
Here, of course, m 1 = m 2 = m s is equal to the mass of the first
particle, w 4 = m s =m 6 equals the mass of the second particle, etc.
* In the first reading, the subsequent derivation up to equation (2.18)
need not be studied in detail.
20 MECHANICS [Part I
Let us multiply both sides of this equation by ~ and sum from 1
to n over i.
Let us first consider the righthand side. We obviously have
in accordance with the law for differentiating composite functions.
For the forces of reaction we obtain
since this equality is a special case of (2.8), in which the displace
ments dxi are taken for all constants q except for # Y ; that is why we
retain the designation of the partial derivative ~ . It is clear that
in such a special displacement the work done by the reaction forces
of the constraints is equal to zero, as in the case of a general displace
ment.
After multiplication by  and summation, the lefthand side
of equality (2.9) can be written in a more compact form, without
resorting to explicit Cartesian coordinates. It is precisely the purpose
of this section to give such an improved notation. To do this we ex
press the kinetic energy in terms of generalized coordinates:
11 11
Substituting the generalized velocities by using (2.5), we obtain
The summation indices for q must, of course, be denoted by different
letters, since they independently take all values from 1 to v inclusive.
Changing the order of summation for Cartesian and generalized co
ordinates we have
n
Tl . v~T fix* dxt ,~ , ~ N
~TftZ>15;ri < 2  13 >
Henceforth, T will have to be differentiated both with respect to
generalized coordinates q a and generalized velocities q a . The co
ordinate q a and its corresponding generalized velocity # a are in
Sec. 2] LAGKANGE'S EQUATION 21
dependent of each other since, in the given position in which the
coordinate has a given value # , it is possible to impart to the system
an arbitrary velocity </ a permitted by the constraints. Naturally, q a
<) T
and </p^ are also independent. It follows that in calculating
all the remaining velocities g^^a. and all the coordinates, including
<7 a , should be regarded as constant.
d T
Let us calculate the derivative x . In the double summation
3q r
(2.13), the quantity y can be taken as the index a and also the index (3,
so that we obtain
8 T 1 . 1 e  r < )xi l 1 d Sj
Both these sums are the same except that in the first the index
is denoted by [3 and in the second by a. They can be combined,
replacing (3 by a in the first summation ; naturally the value of the
sum does not change due to renaming of the summation sign. Then
we obtain
n
^ /rt  ..
(2.14)
v x
Let us calculate the total derivative of this quantity with respect
to time:
d ^T
Here we have had to write down the derivatives of each of the three
factors of all the terms in the summation (2.14) separately.
8 T
Now let us calculate the partial derivative  . As has been
d<? Y
o rri
shown, g a , q& are regarded as constants. Like ^r~, the derivative
COf
dT
Q consists of two terms which may be amalgamated into one.
Differentiating (2.13), we obtain
3T . . VT 8 2 xi dxi /0 ,.
fl  == 9 r a 3& \ m i^  5  7T (2.16)
d ww ^ v ;
22 MECHANICS [Part 1
Subtracting (2.16) from (2.15), we see that (2.16) and the last term
of (2.15) cancel. As a result we obtain
%i /0 im
(2J7)
However, the expression on the righthand side of (2.17) can
dX'
also be obtained from (2.9) if we multiply its lefthand side by ~
and sum over i. For this reason, (2.17), in accordance with (2.10),
Q JJ
is equal to . Thus we find
^ dg Y
d dT _ W __3U^
dt dqf c)</ Y dqf ' \ 1 * )
In mechanics it is usual to consider interaction forces that are
independent of particle velocities. In this case U does not involve #,
so that (2.18) may be rewritten in the following form:
The difference between the kinetic and potential energy is called
the Lagrangian function (or, simply, Lagrangian) and is denoted by
the letter L:
L^TU. (2.20)
Thus we have arrived at a system of v equations with v independent
quantities g a , the number of which is equal to the number of degrees
of freedom of the system:
d d L d L A ^ / o o i \
j, a"  a ~=v, l^a<v. (2.21)
dt dqa. dqoi '
These equations are called Lagrange's equations. Naturally, in
(2.21) L is considered to be expressed solely in terms of </ a and </ a ,
the Cartesian coordinates being excluded. It turns out that this
type of equation holds also in cases when the forces depend on the
velocities (see Sec. 21).*
The rules for forming Lagrangc's equations. Since the derivation
of equations (2.21) from Newton's Second Law is not readily evident
we will give the order of operations which, for this given system,
lead to the Lagrange equations.
* In this case, the Lagrangian function does not have the form of (2.20),
where U is a function of generalized coordinates only. However, the form of
equations (2.21) is still valid.
Sec. 2] LAGRANGE'S EQUATION 23
1) The Cartesian coordinates are expressed in terms of generalized
coordinates :
2) The Cartesian velocity components are expressed in terms of
generalized velocities:
3) The coordinates are substituted in the expression for potential
energy so that it is defined in relation to generalized coordinates:
U=U (q l9 . . ., g, ., #v).
4) The velocities are substituted in the expression for kinetic
energy
 2 ^'".~, ,
which is now a function of # a and g a . It is essential that in generalized
coordinates, T is a function both of # and </ a .
5) The partial derivatives TTT and ^ are found.
dq& dq<x.
6) Lagrange's equations (2.21) are formed according to the number
of degrees of freedom.
In the next section we will consider some examples in forming
Lagrange's equations.
Exercises
1) Write down Lagrange's equation, where the Lagrangian function has
the form:
2) A point moves in a vortical plane along a given curve in a gravitational
field. The equation of the curve in parametric form is x = x (s), z = z (a). Write
down Lagrange's equations.
The velocities are
dx . , . . dz . , .
x = _ 8 ^ x 8 z _ 8 _= z 8
ds ds
The Lagrangian has the form:
i = y las' 1 + *' 1 )i 1 %*()
Lagrango's equation is
~ m [(x'* + 2 /2 ) ] m s* (x'x" + z'z") + mgz' = .
24 MECHANICS [Part I
Sec. 3. Examples ol Lagrange's Equations
Central forces. Central forces is the name given to those whose
directions are along the lines joining the particles and which depend
only on the distances between them. Corresponding to such forces,
there is always a potential energy, U, dependent on these distances.
As an example, we consider the motion of a particle relative to a
fixed centre and attracting it according to Newton's law. We shall
show how to find the potential energy in this case by proceeding
from the expression for a gravitational force.
Gravitational force is known to be inversely proportional to the
square of the distance between the particles and is directed along
the line joining them:
F a  2  r . (3.1)
r 2 r x '
Here a is the factor of proportionality which we will not define more
precisely at this point, r is the distance between the particles, and
is a unit vector. The minus sign signifies that the particles attract
each other, so that the force is in the opposite direction to the radius
vector r. According to (3.1), the attractiveforce component along x
is equal to
since x is a component of r. But r=V# 2 + ?/ 2 + 2 2 > so
and similarly for the two other component forces. Comparing (3.3)
and (2.7), we see that in the given case
I/ = y. (3.4)
W T e note that the potential energy U is chosen here in such a way
that U (oo) = when the particles are separated by an infinite distance.
The choice of the arbitrary constant in the potential energy is called
its gauge. In this case it is convenient to choose this constant so that
the potential energy tends to zero at infinity.
It is obvious that an expression similar to (3.4) is obtained for
two electrically charged particles interacting in accordance with
Coulomb's law.
Spherical coordinates. Formula (3.4) suggests that in this instance
it is best to choose precisely r as a generalized coordinate. In other
words, we must transform from Cartesian to spherical coordinates.
The relationship between Cartesian and spherical coordinates is
Sec. 3]
EXAMPLES OF LAGRANGE'S EQUATIONS
shown in Fig. 1. The zaxis is called the polar axis of the spherical
coordinate system. The angle & between the radius vector and the
polar axis is called the polar angle; it is complementary (to 90)
to the "latitude." Finally, the angle 9 is analogous to the "longitude"
and is called the azimuth. It measures the dihedral angle between
the plane zOx and the plane passing through the polar axis and the
given point.
Let us find the formulae for the transformation from Cartesian
to spherical coordinates. From Fig. 1 it is clear that
z = rcos&. (3.5)
The projection p of the radius vector onto the plane xOy is
Whence,
p=r sin <0.
x = p cos 9 = r sin & cos 9 ,
y = p sin 9 =r sin sin 9 .
(3.6)
(3.7)
(3.8)
We will now find an expression for the kinetic energy in spherical
coordinates. This can be done either by a simple geometrical con
struction or by calculation according to the method of Sec. 2.
Fig. 1
Although the construction is simpler, let us first follow the compu
tation procedure in order to illustrate the general method. We have :
z = cos
x = r sin
= r sin
'r sn ft,
cos 9 + r cos
sin + ^ cos
cos 9
sin
r sn & sn 99,
sin & cos 99.
Squaring these equations and adding, we obtain, after very simple
manipulations, the following:
26 MECHANICS [Part I
T= vm(i 2 + y 2 + z 2 ) = T (r* + r 2 & 2 + r 2 sin 2 <p 2 ) . (3.9)
2t &
The same is clear from the construction shown in Fig. 2. An arbitrary
displacement of the point can be resolved into three mutually per
pendicular displacements: dr, rd$ and pe&p r sin ^9. Whence
sin 2 dy*. (3.10)
Since the square of the velocity v a =h , (3.9) is obtained from
(3.10) simply by dividing by (dt ) 2 and multiplying by ~ .
Hence, in spherical coordinates, the Lagrangian function is expressed
as
L = 1" ( r 2 + r 2 sin 2 &cp 2 + r* & 2 ) U(r). (3.11)
z
Now in order to write down Lagrange's equations it is sufficient to
calculate the partial derivatives. We have:
OL . dL on dL 2 2 a 
 _ = mr ,  = mr z $ , = mr z sin 2 9 cp ;
'dr V* d? T
ft L > rv 9 , no & U & ] L o t\ r\ 9 & Lt ~
^ = mr sin 2 v 2 4 mr t> 2  ., , ^ = mr 2 sin fr cos 9 cp 2 ,   .
dr T c)r #9 T c?9
These derivatives must be substituted into (2.21), which, however,
we will not now do since the motion we are considering actually reduces
to the plane case (see beginning of Sec. 5).
Twoparticle system. So far we have considered the centre of at
traction as stationary, which corresponds to the assumption of an
infinitely large mass. In the motion of the earth around the sun, or
of an electron in a nuclear field, the mass of the centre of attraction
is indeed large compared with the mass of the attracted particle. But
it may happen that both masses are similar or equal to each other (a
binary star, a neutronproton system, and the like). We shall show
that the problem of the motion of two masses interacting only with
one another can always be easily reduced to a problem of the motion
of a single mass.
Let the mass of the first particle be m l and of the second ra 2 . We
call the radius vectors of these particles, drawn from an arbitrary
origin, r x and r 2 , respectively. The components of r x are x l9 y l9 2 1 ;
the components of r 2 are # 2 , ?/ 2 , z 2 . We now define the radius vector of
the centre of mass of these particles B by the following formula:
j 2 v '
Synonymous terms for the "centre of mass" are the "centre of gravity"
and the "centre of inertia."
Sec. 3] EXAMPLES OF LAGRANGE's EQUATIONS 27
In addition, let us introduce the radius vector of the relative position
of the particles
r = r 1 r 2 (3.13)
Let us now express the kinetic energy in terms of B and r. From
(3.12) and (3.13) we 'have
(3.14)
v '
r 2 =R  . (3.15)
2  v '
The kinetic energy is equal to
Differentiating (3.14) and (3.15) with respect to time and substi
tuting in (3.16), we obtain, after a simple rearrangement,
y = l 2 2 t z _ . 2 (3 17)
2 ^ 2 (m l + m 2 ) \ /
Tf we introduce the Cartesian components of the vectors R (X, Y, Z)
and r (x, y, z), then we obtain an expression for the kinetic energy
in terms of Cartesian components of velocity.
Since no external forces act on the particles, the potential energy
can be a function only of their relative positions: U=U(x,y,z).
Thus, the Lagrangian is
Transition to the centreofmass system. Let us write down La
grange's equations for the coordinates of the centre of mass. We have
Hence, in accordance with (2.21)
X=Y=Z = 0.
These equations can be easily integrated:
X = X Q t + X , Y=Y t+Y , Z=Z t+Z Q , (3.18)
where the letters with the index signify the corresponding values
at the initial time.
28 MECHANICS [Part I
Combining the coordinate equations into one vector equation, we
obtain
Thus, the centre of mass moves uniformly in a straight line quite
independently of the relative motion of the particles.
Reduced mass. If wo now write down Lagrange's equations for
relative motion in accordance with (2.21) the coordinates of the
centre of mass do not appear. It follows that the relative motion
occurs as if it were in accordance with the Lagrangian
(where r 2 = i 2 + ?/ 2 + z 2 ), formed in exactly the same way as the
Lagrangian for a single mass m equal to
(3.20)
.
rn 1 f w 2
This mass is called the reduced mass.
The motion of the centre of mass does not affect the relative motion
of the masses. In particular, we can consider, simply, that the centre
of mass lies at the coordinate origin R = 0.
In the case of central forces (for example, Newtonian forces of attrac
tion) acting between the particles, the potential energy is simply
equal to U (r) [this is taken into consideration in (3.10)], where
f z 2 . Then, if we describe the relative motion in spherical
coordinates, the equations of motion will have the same form as for
a single particle moving relative to a fixed centre of attraction.
The centre of mass can now be considered as fixed, assuming R = 0.
From this, in accordance with (3.14) and (3.15), we obtain the distance
of both masses from the centre of mass:
__ B t _
1 t  ' 2
x f
We see that if one mass is much smaller than the other, w 2 <^ m l9
then r l <^r 2 , i.e., the centre of mass is close to the larger mass. This
is the case for a sunplanet system. At the same time the reduced
mass can also be written thus:
From here it can be seen that it is close to the smaller mass. That is
why the motion of the earth around the sun can be approximately
described as if the sun were stationary and the earth revolved about it
with its own value of mass, independent of the mass of the sun.
Sec. 3]
EXAMPLES OF LAGRANGE'S EQUATIONS
29
Simple and compound pendulums. In concluding this section we
shall derive the Lagrangian for simple and compound pendulums.
The simple plane pendulum is a mass suspended on a flat hinge at
a certain point of a weightless rod of length Z. The hinge restricts the
swing of the pendulum. Let us assume that swinging occurs in the
plane of the paper (Fig. 3). It is clear that such a pend
ulum has only one degree of freedom. We can take
the angle of deflection of the pendulum from the ver
tical 9 as a generalized coordinate. Obviously the veloc
ity of particle m is equal to /<p, so that the kinetic
energy is
m m j 2 . 2
77?*r
The potential energy is determined by the height of
the mass above the mean position z / (1 cos 9). pi<r 3
Whence, the Lagrangian for the pendulum is
L = ~l*y* mgl(l coa<p). (3.22)
A double pendulum can be described in the following way: in mass
m there is another hinge from which another pendulum, which is
forced to oscillate in the same plane (Fig. 4), is suspended. Let the
mass and length of the second pendulum be m l and Z x ,
respectively, and its angle of deflection from the verti
cal, i. The coordinates of the second particle are
Xi = I sin 9 + Z t sin <J> ,
z l = l (1 cos 9) + ^ (1 cos <L) .
Whence we obtain its velocity components:
Zj = / sin 9 9 + Zj_ sin Ad> .
Squaring and adding them we express the kinetic energy
of the second particle in terms of the generalized coordinates 9, <; and
the generalize^ velocities 9, <L :
T l =
2 U l cos (9
The potential energy of the second particle is determined in terms of
z t . Finally, we get an expression of the Lagrangian for a double pen
dulum in the following form:
T
L =
m \
m,
i cos (9 <]>) 9 fy
(m
0039)
(3.23)
30
MECHANICS
[Part I
All the formulae for the Lagrangian functions (3.11), (3.22), and
(3.23) will be required in the sections that follow.
Exercises
1) Writo down Lagrango's equation for an elastically suspended pendulum.
J/t
For such a pendulum, the potential energy of an elastic force is U (Z Z ) 2 ,
2*
whore J is the equivalent length of the unstretched rod and k is a constant,
characteristic of its elasticity.
2) Writo down the kinetic; energy for a system of three particles with masses
m t , w 2 , and m 3 in the form of a sum of the kinetic energy of the centre of mass
and the kinetic energy of relative motion, using the following relative coordi
nates :
Sec. 4. Conservation Laws
The problem of mechanics. If a mechanical system has v degrees
of freedom, then its motion is described by v Lagrangian equations.
Each of these equations is of the second order with respect to the time
derivatives q [see (2.17)]. From general theorems of analysis we
conclude that after integration of this system we obtain 2 v arbitrary
constants. The solution can be represented in the following form:
(4.1)
Differentiating these equations with respect to time, we obtain
expressions for the velocities:
(4.2)
Let us assume that equations (4.1) and (4.2) are solved with respect
to the constants C^, . . . , C^v, so that these values are expressed in
terms of t and q lf . . ., g v , q v . . .,q v .
Then
Sec. 4] CONSERVATION LAWS 31
From the equations (4.3) we see that in any mechanical system
described by 2v secondorder equations there must be 2 v functions of
generalized coordinates, velocities, and time, which remain constant
in the motion. These functions are called integrals of motion.
It is the main aim of mechanics to determine the integrals of
motion.
If the form of the function (4.3) is known for a given mechanical
system, then its numerical value can be determined from the initial
conditions, that is, according to the given values of generalized
coordinates and velocities at the initial instant.
In the preceding section we obtained the socalled centre of mass
integrals R and B (3.18).
Naturally, Lagrange's equations cannot be integrated in general
form for an arbitrary mechanical system. Therefore the problem of
determining the integrals of motion is usually very complicated. But
there are certain important integrals of motion which are given directly
by the form of the Lagrangian. We shall consider these integrals in
the present section.
Energy. The quantity
is called the total energy of a system. Let us calculate its total deriv
ative with respect to time.
We have
___ ___ ___
~dt~  a ~0$7 ~dt "0 "fcT ~J~q^"~ "FfcT ~ ~di" '
The last three terms on the righthand side are the derivatives of the
Lagrangian L, which, in the general form, depend on g, q and t.
In determining $ and its derivative we have made use of the sum
7 f\ T
mation convention. The quantity ^  . in Lagrange's equations
*\ r
can be replaced by = . The result is, therefore,
uC[y.
dff 8L
dr = w < 4  5 >
Consequently, if the Lagrangian does not depend explicitly on time
(O T \
~7j = 01 , the energy is an integral of motion. Let us find the con
ditions for which time does not appear explicitly in the Lagrangian.
If the formulae expressing the generalized coordinates q in terms of
Cartesian coordinates x do not contain time explicitly (which corre
* a is summed from 1 to v (see Sen. 2).
32 MECHANICS [Part 1
spends to constant, timeindependent constraints) then the transfor
mation from x to q cannot introduce time into the Lagrangian.
O TT
Besides, in order that 7 , the external forces must also be
G t
independent of time. When these two conditions constant constraints
and constant external forces are fulfilled, the energy is an integral
of motion. To take a particular case, when no external forces act
on the system its energy is conserved. Such a system is called closed.
When frictioiial forces act inside a closed system, the energy of
macroscopic motion is transformed into the energy of molecular
microscopic motion. The total energy is conserved in this case, too,
though the Lagrangian, which involves only the generalized coordi
nates of macroscopic motion of the system, no longer gives a complete
description of the motion of the system. The mechanical energy of
only macroscopic motion, determined by means of such a Lagrangian,
is not an integral. We will not consider such a system in this section.
Let us now consider the case when our definition of energy (4.4)
coincides with another definition, $=T+U. Let the kinetic energy
be a homogeneous quadratic function of generalized velocities, as
expressed in equation (2.13). For this it is necessary that the con
straints should not involve time explicitly, otherwise equation (2.5)
would have the form
where the partial derivative of the function (2.2) with respect to time
is taken for all constants #. But then terms containing q a in the first
degree would appear in the expression for T.
Since we assume that the potential energy U does not depend on
velocity [see (2.18) and (2.19)], then
8L BT
and the energy is
JP 2* T I 4 \
6<taQq ^> (4.6)
But according to Euler's theorem, the sum of partial derivatives of a
homogeneous quadratic function, multiplied by the corresponding
variables, is equal to twice the value of the function (this can easily
be verified from the function of two variables ax 2 + 2 bxy + cy 2 ).
Thus,
that is, the total energy is equal to the sum of the potential energy
and the kinetic energy, in agreement with the elementary definition.
We note that the definition (4.4) is more useful and general also
in the case when the Lagrangian is not represented as the difference
Sec. 4] CONSERVATION LAWS 33
LT U. Thus, in electrodynamics (Sec. 15) L contains a linear
term in velocity. For the energy integral to exist, only one condition
is necessary and sufficient : ^~ = (if, of course, there are no friction
al forces).
The application of the energy integral to systems with one degree
of freedom. The energy integral allows us, straightway, to reduce
problems of the motion of systems with one degree of freedom to those
of quadrature. Thus, in the pendulum problem considered in the
previous section we can, with the aid of (4.7), write down the energy
integral directly:
_ cos 9 ) m ( 4 S )
The value is determinable from the initial conditions. For example,
let the pendulum initially be deflected at an angle cp and released
without any initial speed. It follows that (p 0. Whence
1 cos 9 ). (4.0)
Substituting this in (4.8), we have
From this, the relationship between the deflection angle and time
is determined by the quadrature
9 . (4.11)
\/cos <p cos 9
Tt is essential that the law governing the oscillation of a pendulum
depend only on the value of the ratio l/g and is independent of the
mass. The integral in (4.11) cannot be evaluated in terms of elementary
functions.
A system in which mechanical energy is conserved is sometimes
called a conservative system. Thus, the energy integral permits reducing
to quadrature the problem of the motion of a conservative system
with one degree of freedom.
Tn a system with several degrees of freedom the energy integral
allows us to reduce the order of the system of differential equations
and, in this way, to simplify the problem of integration.
Generalized momentum. We shall now consider other integrals of
motion which can be found directly with the aid of the Lagrangian.
To do this we shall take advantage of the following, quite obvious,
consequence of Lagrange's equations. If some coordinate </ a does not
appear explicitly in the Lagrangian (^ =0), then in accordance
\ 09* I
with Lagrange's equations
34 MECHANICS [Part I
44^ = 0 (4.12)
dt dq a v '
But then
r\ 7"
p a S3   = const, ( 4 13)
i.e., it is an integral of motion. The quantity p a is called the generalized
momentum corresponding to a generalized coordinate with index a.
This definition includes the momentum in Cartesian coordinates:
pL "I
p x =zmv x = ~ . Summarizing, if a certain generalized coordinate does
vVx
not appear explicitly in the Lagrangian, the generalized momentum
corresponding to it is an integral of motion, i.e., it remains constant
for the motion.
In the preceding section we saw that the coordinates X, F, Z of
the centre of gravity of a system of two particles, not subject to the
action of external forces, do not appear in the Lagrangian. From this
it is evidentthat
Z = P z (4.14)
are constants of motion.
The momentum of a system of particles. The same thing is readily
shown also for a system of N particles. Indeed, for N particles we can
introduce the concept of the centre of mass and the velocity of the
centre of mass by means of the equations:
(4.15a)
(4.15b)
JTm;
i
The velocity of the ith mass relative to the centre of mass is
r/ = rrR (4.16)
(by the theorem of the addition of velocities). The kinetic energy of
the system of particles is
N N
T = ~ JTwn if = 2>, ( V + R) 2 =
fi ;~i
N N N
= y^wi*/ 1 + B 2>;r;' + y2><* a  (4.17)
ll il .'1
Sec. 4] CONSERVATION LAWS 35
However, from (4.15b) and (4.16) it can immediately bo seen that
N
JTraif/^O, by the definition of r,' and R. Therefore, the kinetic
= i
energy of a system of particles can be divided into a sum of two terms :
the kinetic energy of motion of the centre of mass
and the kinetic energy of motion of the mass relative to the centre
of mass
N
The vectors f,' are not independent; as has been shown, they are
N
governed by one vector equation J^ w,r,' = 0. Consequently, they can
be expressed in terms of an N 1 independent quantity by determining
the relative positions of the ith and first masses. For this reason the
kinetic energy of N particles relative to the centre of mass is, in general,
the kinetic energy of their relative motion, and is expressed only in
terms of the relative velocities i^ r,. By definition, no external forces
can act on the masses in a closed system, and the interaction forces
inside the system can be determined only by the relative positions
r t r,.
Thus, only R appears in the Lagrangian, and R does not. Therefore,
the overall momentum is conserved:
ar ( N \
P  ~ = \nii tt = const . (4.18)
Equality (4.18), which contains a derivative with respect to a vector,
should be understood as an abbreviated form of three equations:
T^ dL n 8L o dlj
For more detail about vector derivatives see Sec. 11.
We have seen that the overall momentum of a mechanical system
not subject to any external force is an integral of motion. It is impor
tant that it is what is known as an additive integral of motion, i.e.,
it is obtained by adding the momenta of separate particles.
36 MECHANICS [Part I
It may be noted that the momentum integral exists for any system
in which only internal forces are operative, even though they may be
frictional forces causing a conversion of mechanical energy into
heat.
If we integrate (4.18) with respect to time once again, the result will
be the centreofmass integral similar to (3.18). This will be the so
called second integral (for it contains two constants); it contains only
coordinates but not velocities. (3.18) and (4.11) are also second
integrals.
Properties of the vector product. The angular momentum of a particle
is defined as
MFrp]. (4.19)
Here the brackets denote the vector product of the radius vector of
the particle and its momentum. We know that (4.19) takes the place
of three equations,
M x = ypzzpy, M Y = zp x  ~xp z , M z xp y yp x ,
for the Cartesian components of the vector M.
Recall the geometrical definition for a vector product. We construct
a parallelogram on the vectors r and p. Then [rpj denotes a vector
numerically equal to the area of the parallelogram with direction
perpendicular to its plane. In order to specify the direction of [rp]
uniquely, we must agree on the way of tracing the parallelogram con
tour. We shall agree always to traverse the contour beginning with the
first factor (in this case beginning with r). Then that side of the plane
will be considered positive for which the direction is anticlockwise.
The vector [rp] is along the normal to the positive side of the plane.
In still another way, if we rotate a corkscrew from r to p, then it will
be displaced in the direction of [rp]. The direction of traverse changes
if we interchange the positions of r and p. Therefore, unlike a conven
tional product, the sign of the vector product changes if we inter
change the factors. This can also be seen from the definition of Carte
sian components of angular momentum.
The area of the parallelogram is rp sin a, where a is the angle between
r and p. The product r sin a is the length of a perpendicular drawn
from the origin of the coordinate system to the tangent to the trajec
tory whose direction is the same as p. This length is sometimes called
the "arm" of the moment.
The vector product possesses a distributive property, i.e.,
Hence, a binomial product is calculated in the usual way, but the
order of the factors is taken into account.
[a + d, b + c] = [ab] + [ac] + [db] + [dc] .
Sec. 4] CONSERVATION LAWS 37
The angular momentum of a particle system. The angular momentum
of a system of particles is defined as the sum of the angular momenta
of all the particles taken separately. In doing so we must, of course,
take the radius vectors related to a coordinate origin common to all
the particles:
N
We shall show that the angular momentum of a system can be
separated into the angular momentum relative to the motion of the
particles and the angular momentum of the system as a whole, similar
to the way that it was done for the kinetic energy. To do tin's Ave must
represent the radius vector of each particle as the sum of the radius
vector of its position relative to the centre of mass and the radius
vector of the centre of mass ; we must expand the expression for the
particle velocities in the same way. Thou, the angular momentum can
he written in the form
N
N N N N
11 1=1 l Jl
111 the second and third sums, wo can make use of the distributive
property of a vector product and introduce the summation sign
inside the product sign. However, both these sums are equal to zero,
by definition of the centre of mass. This was used in (4.17) for veloc
ities. Thus, the angular momentum is indeed equal to the sum of the
angular momenta of the centre of mass (M ) and the relative motion
(M'):
M = [RP] + JTfr' Pi'J  M + M'. (4.21)
Let us perform these transformations for the special case of a
system of two particles. We substitute r x and r 2 expressed [from
(3.14) and (3.15)] in terms of r and R. This gives
M = friPil + [r2P 2 ] = [tt,Pi + P 2 ] + (
Further, we replace ^ by m l r 1? p 2 by m z r 2 and P!+P 2 by P, after
which the angular momentum reduces to the required form:
38 MECHANICS [Part I
M = [RP] +  w ^ [r r] = [RPJ + [rp] . (4.22)
Here, " m ^^ r = wr = p is the momentum of relative motion.
We shall now show that the determination of angular momentum
of relative motion does not depend on the choice of the origin. Indeed,
if we displace the origin, then all the quantities r/ change by the same
amount ri'=r," + a.
Accordingly, angular momentum for relative motion will be
N N N
N
because
N N
Thus, the determination of angular momentum for relative motion
does not depend on the choice of the origin of the coordinate
system.
Conservation ol angular momentum. We shall now show that the
angular momentum of a system of particles not acted upon by any
external forces is an integral of motion.
Let us begin with the angular momentum of the system as a whole.
Its time derivative is equal to zero:
because P = for any system not acted upon by external forces, and
R is in the direction of P, so that the vector product [RP] = 0.
We shall now prove that angular momentum is conserved for
relative motion. The total derivative with respect to time is
i  1
Here, the first term on the righthand side is equal to zero since r/
is in the direction of p,'. We consider the second term. Let us choose
the origin to be coincident with any particle, for example, the first.
As a result, M', as we have already seen, does not change. The potential
energy can only depend on the differences r x r 2 , r x r a , . . . , ^ T&, . . .
Sec. 4] CONSERVATION LAWS 39
The other differences are expressed in terms of these, for example,
Yk ?I = (TI r/) (r x r^). We introduce the abbreviations.
p rj t\ JT n jj
Then the derivatives^ , ^ , . . . ,  . . . will be expressed in terms
of the variables p x , . . . , p fe _ i , . . . as follows :
fe1
su su
Substituting this in (4.23) we obtain
dt '
k~l /?=!
N1 n Nl Nl
In this expression, only the relative coordinates p x , ...,PNI
remain. We shall now show that, for a closed system, the righthand
side is identically equal to zero. The potential energy is a scalar function
of coordinates. Hence, it can depend only on the scalar arguments
pfe 2 > P/ 2 > (p/ Pfe)> totally irrespective of whether the initial expression
was a function only of the absolute values  r; r*  , or whether it
also involved scalar products of the form (r rk, r/ r). An essential
point is that the system is closed (in accordance with the definition,
see page 32), and the forces in it are completely defined by the relative
positions of the points and by nothing else. Therefore, the potential
energy depends only on the quantities r, r*, and only in scalar
combinations (r/ rfe, r/ r rt ) (in particular, the subscripts i, I; k, n
can even be the same; then the scalar product becomes the square
of the distance between the particles i, k).
To summarize, the potential energy U depends on the following
arguments:
u= = u Oi> P'> > 9k, , PNI; (Pi P 2 )>   , (pfep/)] 
40 MECHANICS [Part I
In order to save space we will, in future, perform the operation for
two vectors, though this operation can be directly generalized to any
arbitrary number. We obtain
CJU __ dU j)(pf) dU fr(p lPa )
~" "~ ~r ~ '
_ [V = dU _
The partial derivatives of the scalar quantities pj, (p x p 2 ) with respect
to the vector arguments are in the given case easily evaluated. Thus.
Each of these equations is a shortened form of three equations referring
to the components (the components of p, are &, TJ/, &):
Henco,
Substituting (4.25) into (4.24) for the case of the two variables, we
obtain
r/M' or . dU or , dU /r lir ^ 8U
 ir =  2 [ Pl pJ ^.y  2 fp 2 p 2 l.  (W _ ([ Pl pj + [p 2 p J) ^_ p  .
But the sign of a vector product depends on the order of the factors
[Pi p2.1 ^ T?2 Pi] Hence it can also be seen that [ Pl Pl ] = [ Pl Pl ] =
and [pspo]^^ Tlicrefore, , =0, as stated.
The integral, like the angular momentum, can also be formed when
the forces are determined not only by the relative position of the par
ticles but also by their relative velocities. This is the case, for example,
in a system of elementary currents interacting in accordance with the
BiotSavart law.
Additive integrals of motion for closed systems. We have thus shown
that a closed system has the following first integrals of motion:
energy, three components of the momentum vector and three compo
nents of the angularmomentum vector. Momentum and angular
momentum are always additive, while energy is additive only for the
non interacting parts of a system.
All the other integrals of motion are found in a much more compli
cated fashion and depend on the specific form of the system (in the
sense that one cannot give a general rule for their definition).
SeC. 5] MOTION IN A CENTIiAL FIELD 41
Exorcises
Describe the motion of a point moving along a cycloid in a gravitational
field.
The equation of the cycloid in parametric form is
z = R cos s, x=
The kinetic energy of the point is
T=~ (x 2 + z 2 )  2
^
The potential energy is U = rtujR cos s.
The totalenergy integral is
<? = 2 7ft ft cos 2  $ 2 mgR cos 5 = const .
2i
The value $ can be determined on the condition that the velocity is equal
to zero when the deflection is maximum s s ; the particle moves along the
cycloid from that position. Hence,
<f = tug R COS S' .
After separating the variables and integrating, wo obtain
cos s
Calling sin y = u, wo rewrite the integral in the form :
arc sin 
In order to find the period of the motion, we must take the integral between
the limits  M O and \ U Q and double the result. This corresponds to the oscilla
tion of the particle within the limits 8= s arid s = v
Thus the total period of oscillation is equal to 4:c I/
Hence, as long as the particle moves on the cycloid, the period of its oscillation
does not depend on the oscillation amplitude (Huygens' cycloidal pendulum).
The period of oscillation of an ordinary pendulufn, describing an arc of a circle,
is known, in the general case, to depend on the amplitude [see (4.11)].
Sec. 5. Motion in a Central Field
The angularmomentum integral. We shall now consider the motion
of two bodies in a frame of reference fixed in the centre of mass. If
the origin coincides with the centre of mass, then R = 0. As was shown
in the preceding section, the angular momentum of relative motion
is conserved in any closed system; specifically, it is also conserved in
42 MECHANICS [Part I
a system of two particles. If the radius vector of the relative position
of the particles r FJ r 2 , and the momentum of relative motion is
(
' V '
then the angularmomentum integral is reduced to the simple form :
M=[rp]= const. (5.2)
It follows that the velocity vector and the relative position vector
all the time remain perpendicular to the constant vector 31 ; in other
words, the motion takes place in a plane perpendicular to M (Fig. 5).
When transforming to a spherical coordinate
system, it is advisable to choose the polar axis
along M. Then the motion will take place in the
plane xy or 9 = " , sin = 1, <i> = 0.
The potential energy can depend only on the
absolute value r, because this is the only scalar
quantity which can be derived solely from the
vector r. In accordance with (3.9), the Lagran
gian for plane motion, with fr = 0, sin o> = l, is
L=TT(r* + r*<?*)U(r) 9 (5.3)
Fig. 5
where m is the reduced mass.
Angular momentum as a generalized momentum. We shall now show
Q *
that M z M is nothing other than ^p , i.e., the component of angular
momentum along the polar axis is a generalized momentum, provided
the angle of rotation cp around that axis is a generalized coordinate.
Indeed, in accordance with (5.2), the angular momentum M is
M = M z xp y yp x ~ mr cos 9 (r sin 9+7* cos 99 )
mr sin 9 (r cos 9 r sin <py)=mr 2 9 (cos 2 9 + sin 2 9)=wr 2 9.
On the other hand, differentiating L with respect to 9 we see that
o y
P * = ~0T = ^^ = MZ * ( 5<4 )
The expression for angular momentum in polar coordinates can also
be derived geometrically (Fig. 6). In unit time, the radius vector r
moves to the position shown in Fig. 6 by the dashed line. Twice the
area of the sector OAB, multiplied by the mass w, is by definition
equal to the angular momentum [cf. (5.2)]. But, to a first approxi
mation the area of the sector is equal to the product of the modulus
r and $ . The height h is proportional to the angle of rotation in unit
Sec. 5] MOTION IN A CENTRAL FIELD 43
time and to the radius itself so that the area of the sector is 1/2 r 2 9.
Thus, a doubled area multiplied by the mass m is indeed equal to the
angular momentum.
The quantity 1/2 r 2 9 is the socalled areal velocity, or the area
described by the radius vector in unit time. The law of conservation
of angular momentum, if interpreted geometrically,
expresses constancy of areal velocity (Kepler's Second
Law).
The central field. If one of two masses is very much
greater than the other, the centre of mass coincides with
the larger mass (see Sec. 3). In this case, the particle
with the smaller mass moves in the given central "field of
the heavy particle. The potential energy depends only on
the distance between the particles and does not depend
on the angle 9. Then, in accordance with (4.12), p^ M z
is the integral of motion. However, since one particle is
considered at rest, the origin should be chosen coincident
with that particle and not with some arbitrary point, as
in the case of the relative motion of two particles. In
the case of motion in a central field, angular momentum is conserved
only relative to the centre.
Elimination ol the azimuthal velocity component. The angular
momentum integral permits us to reduce the problem of twoparticle
motion, or the problem of motion of a single particle in a central field,
to quadrature. To do this we must express 9 in terms of angular mo
mentum and thus get rid of the superfluous variable, in as much the
angle 9 itself does not appear in the Lagrangian. Such variables,
which do not appear in L, are termed cyclic.
In accordance with (4.7), we first of all have the energy integral
f=~(r* + r*i*)+U(r). (5.5)
Eliminating 9 with the aid of (5.4) we obtain
U(r). (56)
This firstorder differential equation (in r) is later on reduced to
quadrature. Before writing down the quadrature, let us examine it
graphically.
The dependence of the form of the path on the sign of the energy.
For such an examination, we must make certain assumptions about
the variation of potential energy.
From (2.7), force is connected with potential energy by the relation
44 MECHANICS [Part I
The upper limit in the integral can be chosen arbitrarily. If F (r)
00
tends to zero at infinity faster than , then the integral I F dr is
00 f
convergent. Then we can put U (r)^( F dr, or /(oo) 0. In other
words, the potential energy is considered zero at infinity. The choice
of an arbitrary constant in the expression for potential energy is called
its yawje.
In addition we shall consider that at r~0, U (r) does not tend to
infinity more rapidly than  , as, for example, for Newtonian attrac
oo
, . f T I* (t , tt
turn U  dr  .
J r z r
r
Let us now write (5.6) as
The lefthand side of this equation is essentially positive. For /=oo
the last two terms in (5.7) tend to zero. Thus, for the particles to be
able to recede, from each other an infinite distance, the total energy
must be positive when the gauge of the potential energy satisfies
f7(oo)=0.
Given a definite form of U, we can now 7 plot the curve of the function
The index M in U denotes that the potential energy includes the
I/ 2
"centrifugal" energy ./ ^r. The derivative of this quantity with
Af 2
respect to r, taken with the opposite sign, is equal to "". If we put
M ~ wr a 9, the result will be the usual expression for "centrifugal
force." However, henceforth, we will call a mechanical quantity of
different origin the "centrifugal force" (see Sec. 8). Let U <() and
monotonic. Since U (oo) = 0, we see that U (r) is an increasing function
of r. It follows that the force has a negative sign (since F 41 ,
i.e., it is an attractive force. Let us assume, in addition, that at infinity
I W ('*) ! > V M ,2~* Let us summarize the assumptions that we have
made concerning UM (r):
1) U\t(r) is positively infinite at zero, where the centrifugal term
is predominant.
2) at infinity, where U (r) predominates, UM (r) tends to zero from
a negative direction.
Sec. 5] MOTION IN A CENTRAL FIELD 45
Consequently, the curve UM (r) has the form shown in Fig. 7,
since we must go through a minimum in order to pass from a decrease
for small values of r to an increase at large values of r.*
In this figure we can also plot the total energy 6\ But since the total
energy is conserved, the curve of $ must have the form of a horizontal
straight line lying above or below the abscissa, depending on the sign
of 6.
For positive values of energy, the line <? = const lies above the curve
UM (r) everywhere to the right of point A . Hence the difference
$ UM(T) to the right of A is positive.
The particles can approach each other
from infinity and recede from each other
to infinity. Such motion is termed infinite.
As we will see later in this section, in the
case of Newtonian attraction, we obtain
hyperbolic orbits.
For <5"<0, but higher than the mini
mum of the curve UM (r), the difference
&* UM (r), i.e., ^ , remains positive
only between the points B and B' (finite
motion). Thus, between these values of
the radius there is included a physi Fig. 7
cally possible region of motion, to which
there correspond elliptical orbits in the case of Newtonian
attraction. In the case of planetary motion around the sun, point B
is called the perihelion and point B 1 the aphelion.
For ? = () the motion is infinite (parabolic motion).
If U (/)>(>, which corresponds to repulsion, the curve UM(T) does
not possess a minimum. Then finite motion is clearly impossible.
Falling towards the centre. For Newtonian attraction, U (0) tends
to infinity like 1/r. If we suppose that U (0) tends to oo more
rapidly than 1/r 2 , then the curve UM(^) is negative for all r close
to zero. Then, from (5.7), f 2 is positive for infinitesimal values of r
and tends to infinity when r tends to zero. If r < initially, then r
does not change sign and the particles now begin to move towards
collision. In Newtonian attraction this is possible only when the
particles are directed towards each other; then "the arm" of the
angular momentum is equal to zero and, hence, the angular momentum
itself is obviously equal to zero, too, so that UM (r) = U (r). If an
initial "arm" exists within the distance of minimum approach, then
M 2
* Ff I U (r)  < 2" a t infinity, then the curve approaches zero on the
positive side, and there can bo a further small maximum after the minimum.
This form of UM (r) applies to the atoms of elements with medium and large
atomic weights.
46 MECHANICS [Part I
the angular momentum M = mvp ^ (p is the "arm") and the motion
can in no way become radial.
In the case of Newtonian or Coulomb attraction for a particle with
angular momentum not equal to zero, there always exists a distance
M 2
r 'for which r becomes greater than $ U (r ). This distance
determines the perihelion for the approaching particles.
However, if U (r) tends to infinity more rapidly than  ^ then,
as r>0, there will be no point at which UM (r) becomes zero. In place
of a hyperbolic orbit, as in the case of Newtonian attraction, a spiral
curve leading to one particle falling on the other results. The turns
of the spiral diminish, but the speed of rotation increases so that the
angular momentum is conserved, as it should be in any central field.
But the "centrifugal" repulsive force turns out to be less than the
forces of attraction, and the particles approach each other indef
initely.
Of course, the result is the same if the energy is negative (for example,
part of the energy is transferred to some third particle, which then
recedes). In the case of attractive forces increasing more rapidly than
1/r 3 , no counterpart to elliptical orbits exists.
If three bodies in motion are subject to Newtonian attraction, two
of them may collide even if, initially, the motion of these particles
was not purely radial. Indeed, in the case of three bodies, only the
total angular momentum is conserved, and this does not exclude the
collision of two particles.
Reducing to quadrature. Let us now find the equation of the tra
jectory in general form. To do this we must, in (5.6), change from
differentiation with respect to time to differentiation with respect to 9.
Using (5.4) we have
dt = ljd<p. (5.9)
Separating the variables and passing to 9 in (5.6) gives
M dr
9 =
Here, the lower limit of the integral corresponds to 9 = 0. If we cal
culate 9 with respect to the perihelion, then the corresponding value
r = r can be easily found by noting that the radial component of
velocity r changes sign at perihelion (r has a minimum, and so dr = 0).
From this we find the equation for the particle distance at perihelion :
*=?* +U M < 6  n >
Sec. 5] MOTION IN A CENTRAL FIELD 47
Kepler's problem. Thus, the problem of motion in a central field
is reduced to quadrature. The fact that the integral sometimes cannot
be solved in terms of elementary functions is no longer so essential.
Indeed, the solution of the problem in terms of definite integrals
contains all the initial data explicitly; if these data are known, the
integration can be performed in some way or other.
But, naturally, if the integral is expressed in the form of a wellknown
function, the solution can be more easily examined in the general
form. In this sense an explicit solution is of particular interest.
Such a solution can be found in only a few cases. One of these is
the case of a central force diminishing inversely as the square of
distance. The forces of Newtonian attraction between point masses
(or bodies possessing spherical symmetry) are subject to this
law.
It will be recalled that the laws of motion in this case were found
empirically by Kepler before Newton deduced them from the equations
of mechanics and the law of gravitation. It was the agreement of
Newton's results with Kepler's laws that was the first verification of
the truth of Newtonian mechanics. The problem of the motion of a
particle in a field of force diminishing inversely with the square of the
distance from some fixed point, is called Kepler's problem. The prob
lem of the motion of two bodies with arbitrary masses always reduces
to the problem of a single body when passing to a frame of reference
fixed in the centre of mass.
The expression "Kepler's problem" can also be applied to Coulomb
forces acting between point charges. These can either be forces of
attraction or repulsion. In all cases we shall write U = , where a <
for attractive forces and a>0 for repulsive forces.
If we replace in (5.10) by a new variable x, the integral in the
Kepler problem is reduced to the form
rr4
dx + M
. 2a
___ ffi ._ _ /J I
X +
= arc cos
_ ,
M m M*

] a*
If ,
V M*
M
x ~ 
mr
M
x =
At the lower limit, the expression inside the arc cos sign is equal
to unity [as will be seen from (5.11)], since the lower limit was chosen
on the condition that dr = 0. But arc cos 10. Rearranging the result
of integration and reverting to r, we obtain, after simple manipu
lations,
M*
r = * . (6J2 )
. , M ] a* , 2^ V '
1 H tf 5TrT H COS <p
a V M 2 m
48 MECHANICS [Part I
(5.12) represents the standard equation for a conic section, the
eccentricity being equal to I/ 1 4 \ 2 . As long as this expression
is less than unity, the denominator in (5.12) cannot become zero,
because cos cp< 1. But this is true for " <$<(). Thus, when
<^<0, the result is elliptical orbits. For this it is necessary that a<0,
i.e., that there be an attraction, otherwise (5.12) would lead to r<0,
which is senseless.
For ff > the eccentricity is greater than unity and the denominator
in (5.12) becomes zero for a certain 9 = 900. Thus, the orbit goes to
infinity (a hyperbola). The direction of the asymptote is obtained by
putting r =00 in (5. 12). This requires that cos 900 * .
M I/ a 2 2<^
V ~M* + ~w~
The angle between the asymptotes is equal to 2 900, when the particles
repel each other, and to 2 (7^900), when the particles attract. An
example of a trajectory, when the forces are repulsive, is shown in
Fig. 8, Sec. 6.
Exercise
ar 2
Obtain iho equation, of the trajectory when U = $ > 0.
2
See. 6. Collision of Particles
The significance of collision problems. In order to determine the
forces acting between particles, it is necessary to study the motion of
particles caused by these forces. Thus, Newton's gravitational law
was established with the aid of Kepler's laws. Here, the forces were
determined from iinite motion. However, infinite motion can also be
used if one particle can, in some way, be accelerated to a definite
velocity and then made to pass close to another particle. Such a process
is termed 4 'collision" of particles. It is not at all assumed, however,
that the particles actually come into contact in the sense of "collision"
in everyday life.
And neither is it necessary that the incident particle should be
artificially accelerated in a machine: it may be obtained in ejection
from a radioactive nucleus, or as the result of a nuclear reaction, or
it may be a fast particle in cosmic radiation.
Two approaches are possible to problems on particle collisions.
Firstly, it may be only the velocities of the particles long before the
collision (before they begin to interact) that are given, and the problem
is to determine only their velocities (magnitude and direction) after
they have ceased to interact. In other words, only the result of the
collision is obtained without a detailed examination of the process.
In this case, some knowledge of the final state must be available (or
Sec. 6] COLLISION OF P ARTICLES 49
specified) beforehand: it is not possible to determine, from the initial
velocities alone, all the integrals of motion which characterize the
collision, and, hence, it is likewise impossible to predict the final state.
With this approach to collision problems, only the momentum and
energy integrals are known.
However, another approach is possible : it is required to precalculate
the final state where the precise initial state is given.
Let us first consider collisions by the first method. It is clear that
if only the initial velocities of the particles are known, the collision is
not completely determined: it is not known at what distance the par
ticles were when they passed each other. This is why some quantity
relating to the final state of the system must be given. Usually the
problem is stated as follows : the initial velocities of the colliding par
ticles and also the direction of velocity of one of them after the colli
sion are specified. It is required to determine all the remaining quanti
ties after the collision. In such a form the problem is solved uniquely.
Six quantities are unknown, namely the six momentum components
of both particles after the collision. The conservation laws provide
four equalities: conservation of the scalar quantity (energy) and
conservation of the three components of the vector quantity (momen
tum). Therefore, with six unknowns, it is necessary to specify two
quantities which refer to the final state. They are contained in the
determination of the unit vector which specifies the direction of the
velocity of one of the particles ; an arbitrary vector is defined by three
quantities, but a unit vector, obviously, only by two. Actually,
only the angle of deflection of the particles after the collision need be
given, i.e., the angle which the velocity of the particle makes with
the initial direction of the incident particle. The orientation (in
space) of the plane passing through both velocity vectors is im
material.
Elastic and inelastic collisions. A collision is termed elastic, if the
initial kinetic energy is conserved when the particles separate after
the collision at infinity, and inelastic, if, as a result of the collision,
the kinetic energy changes at infinity. In nuclear physics, studies are
very often made of collisions of a more general character, in which the
nature of the colliding particles changes. These collisions are also
inelastic. They are called nuclear reactions.
The laboratory and centreo!mass frames of reference. When colli
sions are studied in the laboratory, one of the particles is usually at
rest prior to collision. The frame of reference fixed in this particle (and
in the laboratory) is termed the laboratory frame. However, it is more
convenient to perform calculations in a frame of reference, relative
to which the centre of mass of both particles is at rest. In accordance
with the law of conservation of the centre of mass (3.18), it will also
be at rest in its own frame after the collision. The velocity of the
centre of mass, relative to the laboratory frame of reference, is
60 MECHANICS [Part I
(6.1)
v '
Here V is the velocity of the first particle (of mass m^ relative to the
second (with mass m 2 ). In so far as the second particle is at rest in
the laboratory system, v is also the velocity of the first particle relative
to this system.
The general case of an inelastic collision. The velocity of the first
particle relative to the centre of mass is, according to the law of
addition of velocities, equal to
(6.2)
v
and in the same system, the velocity of the second particle is
(6.3)
Thus, m x V 10 + m 2 V 2 o = 0, as it should be in the centreofmass
system.
In accordance with (3.17), the energy in the centreofmass system is
(64)
Here, the reduced mass is indicated by a zero subscript, since in nuclear
reactions it may change.
Let the masses of the particles obtained as a result of the reaction
be w s and ra 4 , and the energy absorbed or emitted Q (the socalled
"heat" of the reaction). If Q is the energy released in radiation, then,
strictly speaking, one should take into account the radiated momen
tum (see Sec. 13). But it is negligibly small in comparison with the
momenta of nuclear particles.
Thus, the law of conservation of energy must be written in the follow
ing form:
=+=. (6.5)
Here, m =  ~^ is the reduced mass of the particles produced in
^3 *T~ ^4
the nuclear reaction, and v is their relative velocity.
In order to specify the collision completely, we will consider that
the direction of v is known, since the value of v is determined from
(6.5). Then the velocity of each particle separately will be
They satisfy the requirement m z V 30 +w 4 v 40 = 0, i.e., the law of
conservation of momentum in the centreofmass system, and give
the necessary value for the kinetic energy
Sec. 6] COLLISION OF PABTICLES 51
Now, it is not difficult to revert to the laboratory frame of reference.
The velocities of the particles in this system will be
V  ___
Y
Equations (6.7) give a complete solution to the problem provided
the direction of v is given.
Elastic collisions. The computations are simplified if the collision is
elastic, for then m z = mi, w 4 = w 2 > Q = 0. It follows from (6.5) that
the relative velocity changes only in direction and not in magnitude.
Let us suppose that its angle of deflection x is given. We take the axis
Ox along v , and let the axis Oy lie in the plane of the vectors v and v
(which are equal in magnitude in the case of elastic collisions). Then
V X = VQ COS X, Vy = VQ SHI X
From (6.6), the components of particle velocity in the centreofmass
system will, after collision, be correspondingly equal to
.. _
i 20y
Since the velocity of the centre of mass is in the direction of the axis
Ox, we obtain, from (6.7), the equations for the velocities in the
laboratory frame of reference:
xjtjo m 2 v sinx
Viy ~ V " v ~ m~+m~ '
_ _ m,v sinx
""  ~
  114 + tn, J 2y
By means of these equations, the deflection angle 6 of the first
particle in the case of collision in the laboratory system can be related
to the angle x (i e., its deflection angle in a centreofmass system):
(6.8)
The "recoil" angle of the second particle 0' is defined as
v y sin x 4. X
~^TCOSX~ CO y (69)
6' _*
2 2 '
52 MECHANICS [Part I
The minus sign in the definition of tg 0' is chosen because the signs
of v lY and v 2y are opposite.
The case of equal masses. Equation (6.8) becomes still simpler if the
masses of the colliding particles are equal. This is approximately true
in the case of a collision between a neutron and proton. Then, from
(6.8).
tan 6 tan , 0>
i.e., the particles fly off at right angles and the deflection angle of the
neutron in the laboratory system is equal to half the deflection angle
in the ceritreofmass system. Since the latter varies from to 180,
cannot exceed 90. And, in addition, the velocity of the incident
particle is plotted as the "resultant" velocity of the diverging particles.
The collision of billiard balls resembles the collision considered here
of particles of equal mass, provided that the rotation of the balls
about their axes is neglected.
The energy transferred in an elastic collision. The energy received
by the second particle in a collision is
JP __ m z ml (lcosx) vj
(mi + "!,)
Its portion, relative to the initial energy of the first particle, is
<? 2 __ 2m t w 2 (1 cos/)
\2
(6.10)
& 7
From this we obtain ~~ = sin 2 =sin 2 for particles of equal
.
mass. Accordingly, the portion of the energy retained by the first
particle is  ==cos 2 0. In a "headon" collision x^ 18( ) , 090.
0Q
The first particle comes to rest and the second continues to move
forward with the same velocity. This can easily be seen when billiard
balls collide.
The problem of scattering. Let us now examine the problem of colli
sion in more detail. We shall confine ourselves to the case of elastic
collision and perform the calculations in the centreofmass system.
The transformation to the laboratory system by equations (6.7) is
elementary.
It is obvious that for a complete solution of the collision problem,
one must know the potential energy of interaction between the par
ticles U (r) and specify the initial conditions, so that all the integrals
of motion may be determined. The angularmomentum integral is
found in the following way. Fig. 8. which refers to repulsive forces,
Sec. 6] COLLISION OF PARTICLES 53
represents the motion of the first particle relative to the second.
The path at an infinite distance is linear, because no forces act between
the particles at infinity. Since the path is linear at infinity, it possesses
asymptotes. The asymptote for the part of the trajectory over which
the particles are approaching is represented
by the straight line Al\ and FB is the
asymptote for the part where they recede.
The collision parameter. The distance of
an asymptote from the straight line OG,
drawn through the second point and parallel
to the relative velocity of the particles at
infinity, is called the collision parameter
f'aimmg distance"). It has been denoted
by p, since, as can be seen from Fig. 8, p Ls
also the "arm" of the angular momentum.
If there were 110 interaction between the particles, they would pass
each other at a distance p ; this is why p is called the collision parameter.
But we know that the angular momentum is very simply expressed
in terms of p. In the preceding section it was shown that it is equal
to ravp. Let us draw the radius vector OA to some very distant point A.
Then the angular momentum is
M = mvr sin a
(the angle a is shown on the diagram). But rsina = p, so that
(6.11)
Recall that here ra is the reduced mass of the particles and v is their
relative velocity at infinity.
The energy integral is expressed in terms of the velocity at infinity
thus :
since U (oo) = 0.
The deflection angle. The deflection angle x is equal to  re 29 ,,  ,
where 9 CJ is half the angle between the asymptotes. The angle 9
corresponds to a rotation of the radius vector from the position OA,
where it is infinite, to a position OF, where it is a minimum. Hence,
from equation (5.10) the angle 900 is expressed as
oo
/
M dr , 1Q v
*n7aT "Tr^fT 7 "  Trr T7o =" v > (Q.Lt)
r is determined from (5.11). In place of M and $ we must substitute
into (6.13) the expressions (6.11) and (6.12).
54
MECHANICS
[Part I
The differential effective scattering crosssection. Let us suppose
that the integral (6.13) lias been calculated. Then 9^, and therefore x>
are known as functions of the collision parameter p. Let this relation
ship be inverted, i.e., p is determined as a function of the deflection
angle :
P = P(X) (6 14 )
In collision experiments, the collision parameter is never defined
in practice; a parallel beam of scattering particles is directed with
identical velocity at some kind of substance, the atoms or nuclei
of which are scatterers. The distribution of particles as to deflection
angles x ( or > more exactly, as to angles in the laboratory system)
is observed. Thus a scattering experiment is, as it were, performed
very many times one after the other with the widest range of aiming
distances.
Let one particle pass through a square centimetre of surface
of the scattering substance. Then, in an anuulus contained between
p and pMp, there pass 2?rp dp particles. We classify the collisions
according to the aiming distances, similar to the way that it is done
on a shooting target with the aid of a concentric system of rings.
If p is known in relation to x> then it may be stated that dcr= 2;rp dp =
= 27rp~ dx particles will be deflected at the angle between x
and
Let us suppose that the scattered particles are in some way detected
at a large distance from the scattering medium. Then the whole
scatterer can be considered as a point and we can say that after
scattering the particles move in straight lines
from a common centre. Let us consider those
particles which occupy the space between
two cones that have the same apex and a
common axis ; the half angle of the inner cone
is equal to x> an d the external cone x + ^X
The space between the two cones is called a
solid angle, similar to the way that the plane
contained between two straight lines is called
a plane angle. The measure of a plane angle is
the arc of a circle of unit radius drawn about
the vertex of the angle, while the measure
of a solid angle is the area of a sphere of unit radius drawn about the
centre of the cone. An elementary solid angle is shown, in Fig. 9 as
that part of the surface of a sphere covered by an element of arc r/x
when it is rotated about the radius OC. Since OC=\ y the radius
of rotation of the element dx is equal to sin x Therefore, the surface
of the sphere which it covers is equal to 2rc sin xdx* Thus, the elemen
tary solid angle is
Fig. 9
Sec. 6] COLLISION OF PARTICLES 55
(6.15)
The number of particles scattered in the element of solid angle is,
thus,
d a==p *.J*<L. (6.16)
^ d x sin x V '
The quantity da has the dimensions of area. It is the area in which
a particle must fall in order to be scattered within the solidangle
element dl. It is called the effective differential scattering crosssection
in the element of solid angle dQ.
Experimentally we determine just this value, because it is the
angular distribution of the scattered particles that is dealt with
[in (6.16) we consider that p is given in relation to xl If there are n
scatterers in unit volume of the scattering substance, then the attenu
ation of the primary beam J in passing through unit thickness of
the substance, due to scattering in an elementary solid angle dii, is
d Jn = Jndc= Jnp ~ . particles/cm.
If we examine da as a function of x, we find a relationship between
the collision parameter and the deflection angle. And this allows
us to draw certain conclusions about the nature of the forces acting
between the particle and the scattering centre.
Rutherford's formula. A marvellous example of the determination
of forces from the scattering law is given by the classical experiments
of Rutherford with alpha particles. As was pointed out in Sec. 3,
the Coulomb potential acting on particles decreases with distance
according to a law, in the same way as the Newtonian potential.
Consequently, the deflection angle can be calculated from the equations
of Sec. 5. Let us first of all find the angle <p co . It can be determined
from equation (5.12) by putting r = oo, a > (the charges on the nucleus
and alpha particle are like charges). Hence,
The integrals of motion S and M are determined with the aid of
(6.11) and (6.12). We therefore have
(6.18)
I since 9 = * f I . We now form the equation for the effective di
ential scattering crosssection in the centreofmass svstear wt
the aid of (6.16):
66 MECHANICS [Part I
*.. (6.19)
If the scattering nucleus is not too light, this equation, to a good
approximation, also holds in the laboratory system.
Thus, the number of particles scattered in the elementary solid
angle dl = 2n sin x ^X> * s inversely proportional to the fourth power
of the sine of the deflection halfangle. This law is uniquely related
to the Coulomb nature of the forces between the particles.
Studying the scattering of alpha particles by atoms, Rutherford
showed that the law (6.19) is true for angles up to y, corresponding
to collision parameters less than 10~ 12 cm. It was thus experimentally
proved that the whole mass of the atom is concentrated in an ex
ceedingly small region (recall that the sixe of an atom is ^lO" 8 cm.).
Thus, experiments on the scattering of alpha particles led to the
discovery of the atomic nucleus and to an estimation of the order
of magnitude of its dimensions.
Isotropic scattering. As may be seen from equation (6.19), the
scattering has a pronounced maximum for small deflection angles.
This maximum relates to large aiming distances since particles passing
each other at these distances are weakly deflected, while large distances
predominate since they dciiiie a larger area. Thus, if the interaction
force between the particles does not identically convert to zero at
a finite distance, then, for small deflection angles, the expression
for c/a will always have a maximum. This maximum is the more
pronounced, the more rapidly the interaction force decreases with
distance, for in the case of a rapidly diminishing force, large aiming
distances correspond to very small deflection angles.
However, particles that are very little deflected can in no way
be detected experimentally as deflected particles. Indeed, the initial
beam cannot be made ideally parallel. For this reason, when investi
gating a scattered beam, one must always neglect those angles which
are comparable with the angular deviation of the particles in the
initial beam from ideal parallelism.
For a sufficiently rapid attenuation of force with distance, the region
of the maximum of da in relation to the angle % can refer to such
small angles that the particles travelling within these angles will
not be distinguished as being scattered because of their small deflection
angles. On the other hand, the remaining particles will be the more
uniformly distributed as to scattering angle, the more rapidly the
forces fall off with distance.
This can be seen in the example of particles scattered by an im
permeable sphere (Exercise 1). Such a sphere may be regarded as
the limiting case of a force centre repulsing particles according to
the law UJir) = U 1 J ; when n tends to infinity : if r < r , then U (r)>oo,
Sec. 7] SMALL OSCILLATIONS 57
and if r >r , then U (r)>0. When n== oo the scattering is completely
isotropic. If n is large, the angular distribution of the particle is
almost isotropic, and only for very small deflection angles has the
distribution a sharp maximum. Hence, a scattering law that is almost
isotropic indicates a rapid diminution of force with distance.
The scattering of neutrons by protons in the centreofmass system
is isotropic up to energy values greater than 10 Mev (1 Mev equals
1.6 x 10~ 6 erg). An analysis of the effective cross section shows that
nuclear forces are shortrange forces; they are very great at close
distances and rapidly diminish to zero at distances larger than
2 x 10~ 13 cm. It must be mentioned, however, that a correct investi
gation of this case is only possible on the basis of the quantum theory
of scattering (Sec. 37).
Exercises
1) Find tho differential effective scattering cross section for particles by
an impermeable sphere of radius r .
The impermeable sphere can be represented by giving tho potential energy
in tho form U (r) = for r > r (outside the sphere) and U (r)^oo for r< r
(inside tho sphere). Then, whatever tho kinetic energy of tho particle, penetra
tion into the region r <r Q is impossible.
In reflection from the sphere, the tangential component of momentum is
conserved and the normal component changes sign. Tho absolute value of the
momentum is conserved since tho scattering is elastic. A simple construction
shows m that the collision parameter is related to the deflection angle by
p r cos ,
if p < r Q . Hence the general equation gives
so that the scattering occurs isotropically for all angles. Tho total effective
scattering crosssection a is equal, in this case, to 7cr 2 , as expected. Note that
if the interaction converted to zero not at a finite distance but at infinity, the
total scattering crosssection would tend to infinity since to any arbitrarily
large approach distance p there would correspond a certain deflection angle,
and the integral 2 :rp dp diverges. In the quantum theory of scattering,
a is also finite when the forces diminish fast enough with distance.
2) Tho collision of particles with masses m l and ra 2 is considered (the mass
of the incident particle is mj. As a result of the collision, particles with the
same masses are obtained whose paths make certain angles 9 and >\> with the
initial flight direction of the particle of mass m. Determine the energy Q
which is absorbed or emitted in the collision.
See. 7. Small Oscillations
In applications of mechanics, we very often meet a special form
of motion known as small oscillations. We devote a separate section
to the theory of small oscillations.
58
MECHANICS
[Part I
The definition of small oscillations of a pendulum. In the problem
of pendulum oscillations in Sec. 4 it was shown that the equation
relating the deflection angle 9 to time led, in the general case, to
a nonelementary (elliptical) integral (4.11). A simple graphical in
vestigation shows that the function 9 (t) is
periodic. Fig. 10 shows the curve U (<p)=mgl
(1 cos 9), which gives the relationship be
tween potential energy and deflection angle.
The horizontal straight line corresponds to
a certain constant value of S. If $ < 2 mgl,
the motion occurs periodically with time
between the points 9 and 9 .
The problem is greatly simplified if
90^!* i e > ^ e angle 9 is small in com
parison with a radian. Then cos 9 can
be replaced by the expansion 1 ?". Since 9<9 cos 9 can a l so
be replaced by 1  . After this the integral (4.11) can be easily
evaluated :
lo
1/J f rf 9 I
= V . ==11
r 9 J >  2 ' ^
9
arc cos .
(7.1)
<Po
Inverting relation (7.1), we get the angle as a function of time
*. (7.2)
The result is a periodic function. As can be seen from (7.2), the
period 9 is equal to 2n J/ . The quantity I/ is called the frequency
of oscillation
(7.3)
This quantity gives the number of radians by which the argument
of the cosine in (7.2) changes in one second. Sometimes the term
frequency denotes a quantity that is 2 re times less and equal to the
number of oscillations performed by the pendulum in one second.
The inverse value ^ , is the period of small oscillations of the pen
dulum. An important point is that the period and the frequency
of small oscillations do not depend on the amplitude of oscillation 9 .
The general problem of small oscillations with one degree of freedom.
In order to solve the smalloscillation problem, we need not, initially,
reduce to quadrature the problem of arbitrary oscillations; we can
first perform an appropriate simplification of the Lagrangian.
Sec. 7] SMALL OSCILLATIONS 59
First of all, we note that any oscillations, both large and small, always
occur about a position of equilibrium. Thus, a pendulum oscillates
about a vertical. On deflection from the position of stable equilibrium,
a restoring force acts on the system in the opposite direction to the
deflection. In the equilibrium position, this force obviously becomes
zero (by definition of the "equilibrium" concept).
Force is equal to the derivative of potential energy with respect
to the coordinate taken with opposite sign. The equilibrium condition
written in terms of this derivative is
Let us denote the solution of this equation by q = q  We assume,
initially, that the system has only one degree of freedom and expand U
in a Taylor series in the vicinity of the point q Q :
U(q)U (q ) + (qq ) + ? (gj.)' + . . . (7.5)
The linear term relative to q q vanishes in accordance with (7.4).
We denote (05 ) ^J the letter (3. Then, confining ourselves to these
terms of the series, we obtain
U(<D^U( q ) + t(qq )*. (7.6)
The force near the equilibrium position is
F to) = ~ *dT = P to ?o)  (7.7)
For this force to be a restoring force (i.e., for it to act in the direction
opposite to the deflection), the following inequality must hold:
This is the stability condition for the equilibrium ; the function U (q)
must increase on both sides of the point q=q Q , It follows that the
potential energy at that point must be a minimum. This is shown
in Fig. 10 at 9 = 0.
Let us now examine the expression for kinetic energy. If, in the
general formula for the kinetic energy of a particle,
we put x = x(q), y = y(q), z = z(q), then T reduces to the form
.
+ (dq
60 MECHANICS [Part I
The quantity in the brackets depends only on </; and so the kinetic
energy of a particle can be represented in the form
T= Ja( ? )j. (7.9)
Let us now expand the coefficient a (q) in a series, in terms of
q~ q Qf in the vicinity of the equilibrium position:
In order that the particle should not move far from the equilibrium
position, its velocity must be small. In other words, the zero member
of the kinetic energy expansion  t>  a (q l} ) q 2 is already of the same order
of smallness for small oscillations as tho second term in the expansion
Q
of U ((/), i.e., ^(q </ ) 2 . When q^q all the energy of oscillation
is kinetic, while for maximum deflection all the energy is potential.
i A
Therefore ^ a (q ) q* and (q <? ) 2 are of the same order of magnitude,
and the remaining terms in the series [including those containing
(q f/ )</ 2 ] can be neglected. We shall show that the mean values
of both the quantities .^a (</ ) q 2 and  (q q Q ) 2 are the same after
we determine q as an explicit function of time.
In future, the coordinate q will be measured from the equilibrium
position, i.e., we shall put </ =^0. Then [omitting U (0), which does
riot affect the equations of motion] the Lagrangian can be written
in the following form:
Thus, Lagraiige's equation will be written as
oc(0) + pgr = 0. (7.11)
Denoting
w  a(0) a(0) ' v>>i "'
we reduce (7.12) to the general form for the oscillation equation:
> 2 g = 0. (7.13)
Various forms for the solutions of smalloscillation problems. The
general solution of this equation, which contains two arbitrary
constants, may be written in one of three forms:
Sec. 7] SMALL OSCILLATIONS 61
q = C l cos co t + C sin co, (7.14a)
g = (7cos(co + y), (7.14b)
7 = Re{C"c fwl }. (7.14c)
The symbol Ke{} signifies the real part of the expression inside the
braces. The constant C" inside the braces is complex: C' = C l iC 2 .
The constants are chosen in accordance with the initial conditions.
The constant y is called the initial phase, and C is the amplitude.
If we are only interested in the frequency of small oscillations,
and not the phase or amplitude, it is sufficient to use equation (7.12),
verifying that the second derivative I Va1 is positive.
A system which is described by equation (7.13) is called a linear
harmonic oscillator.
It can be seen from equations (7.10), (7.12), and (7.14b) that the
averages of the potential energy and kinetic energy of the oscillator
during one period are the same because the averages of the squares
of a sine or cosine are equal to one half:
sin* (<o* + Y)  sin 2 (a* + y) dt =  ; cos* (at + y) =
Small oscillations with two degrees of freedom. We shall now con
sider oscillations with several degrees of freedom. As an example,
Jet us first take the double pendulum of Sec. 3. If we confine ourselves
to small oscillations, we must consider that the deflections 9 and <p
are close to zero, (i.e., the pendulum is close to a vertical position).
Then, by substituting the equilibrium values of the coordinates
9 and <];, cos (9 ^) in the kinetic energy must be replaced by cos = 1
as in the problem of oscillations with one degree of freedom where
cos 9 and cos <L, in the expression for potential energy, must be re
2 t 2
placed by 1   and 1  ^ . Then the Lagrangian will have the form
L = , V + L Hy + mM ty _ _ i. lg(?2 _ ^
(7.15)
Let us examine this in a somewhat more general form:
U (0)
62 MECHANICS [Part I
Here, the coefficients a n , a 12 , and a 22 are assumed to be constant
numbers expressed in terms of the equilibrium values g x and q 2 .
Comparing (7.15) and (7.16), we find that in the problem of the double
pendulum
a n = (m + raj) I 2 , a 12 = m l ll 1 , a 22 = m l l\ ;
In the general case, the coefficients p n . p 12 and (3 22 are expressed
by the equations
where the derivatives are also taken in the equilibrium positions.
For the equilibrium to be stable, we must demand that the following
inequality be satisfied:
^(?) ^(0) = Y(PiiJ + 2p 12 ? l g t + p aag ;)>0. (7.17)
Under this condition, U has a minimum at the point ^ = 0, q^ = 0.
Let us rewrite the lefthand side of (7.17) in identical form
This expression remains positive for all values of q l and q 2 , provided
the coefficients of both quadratics in q are greater than zero:
Pn>0, (7.18)
(7.19)
In future, we shall consider that the conditions (7.18) and (7.19),
together with analogous conditions for a n , a 12 , and oc 22 , are satisfied.
We shall now write down Lagrange's equations. We have
Whence
a a r = 1 (7
In order to satisfy these equations, we shall look for a solution
in the form
As in (7.14c), the real part of the solution (7.21) must be taken.
Sec. 7] SMALL OSCILLATIONS 63
The equation for frequency. Substituting (7.21) in (7.20), we
obtain equations relating A 1 and A z :
" ' * \ (7.22)
/ Q *,*x2\/* l/n ..9.\yl /^l ^ '
(p 12 a 12 co }
Transferring terms in A 2 to the righthand side of the equation
and dividing one equation by the other, we eliminate A l and A%:
Reducing (7.23) to a common denominator, we arrive at the bi
quadratic equation
(<x n a 22 af a ) to 4 ( P n a 2 2 + 22*11 2a 12 (i 12 ) to 2 +
+ PiiP22 Ma = 0. (7.24)
Substituting here the expressions for a,*, (3ifc from (7.15), we obtain
an equation for the frequencies of a double pendulum
If we introduce still another contraction in notation (for the given
problem) ~ \  = fji, the expression for frequencies will be of
the following form:
It is easy to see that this expression yields only the real values
of the frequencies. However, we shall show this in more general
form for equation (7.24). Let us assume that the following function
is given:
F ( co 2 ) = (a u a 22 of,) co* ( p n a 22 + p 22 a n 2p 12 a 12 ) <* + PiiP 22 Pfi >
which passes through zero for all values of G> that satisfy equation
(7.24). F (co 2 ) is positive for to 2 = and for to 2 = 00, since P n p 22
Pi2 > > a n a 22 a i2 > ^ Let us now substitute into this function
Q
the positive number co 2 =~ 1 . After a simple rearrangement we
obtain
P22  Pl2 22 ) 2 < .
Thus, as co 2 varies from to oo, F (co 2 ) is first positive, then negative,
and then again positive. Hence, it changes sign twice, so that equation
(7.24) has two positive roots coj, cojj and, as was asserted, all the
values for frequency are real.
64 MECHANICS [Part I
The quantity co has four values, both pairs of which are equal
in absolute value. If we represent the solution in the form (7.21),
it is sufficient to take only positive G>.
Normal coordinates. Let us put these roots in (1.22). To each of
^
them there will correspond a definite ratio of the coefficients ~ . For
A i
i I, 2 we have
n
'
A "
Pl2
According to (7.23), the same ratio is also obtained from the second
equation of (7.22). For example, for the double pendulum
6i?
;= //^A ? ; * i' s equal to 1 or 2 depending on the sign in front
of the root in the solution for co 2 .
Each frequency co; defines one partial solution of the system
(7.20). Since the system is linear, the general solution is the sum
of these particular solutions. Let us write this as
*' . (7.26)
We must, of course, take only the real parts of the expressions on
the right.
We now introduce the following notation:
4(JVi' == Qi , A<? *<**' = Qt . (7.27)
According to (7.27), the quantities Qi and Q 2 satisfy the differential
equations
Qj + co^O; (3 2 + col9 2 0. (7.28)
Each of these equations can be obtained from the Lagrangiaii
Lt=*2Q1l'<*1<K, (7.29)
which describes oscillation with one degree of freedom.
Thus, in terms of the variables Qi, the problem of two related
oscillations with two degrees of freedom q^ q 2 has been reduced
to the problem of two independent harmonic oscillations with one
degree of freedom Q l and Q 2 . The coordinates Q l and Q 2 are termed
normal.
In equations (7.20), we cannot arbitrarily put 3 1 = or g 2 = 0:
if the quantity q l oscillates, then it must cause q 2 to oscillate. In
contrast, the oscillations of the quantities Q l and Q 2 are in no way
related [as long as we limit ourselves to the expansion (7.16) for L].
Sec. 7] SMALL OSCILLATIONS 65
From equations (7.26), we can express Qi and Q% i* 1 terms of q l and q 2 :
If, for example, we choose the initial values of q and q so that at
this instant Q = and Q = 0, then the oscillation with frequency G^
will riot occur at all. For this it is sufficient, at t = 0, to take the co
ordinates and velocities in a relationship such that 2 9i #2 ==
and 2 i ?2 == ^ I n ther words, only the frequency co 2 will occur,
and the oscillations will be strictly periodic. When both frequencies
Oj and co 2 are excited they are generally speaking incommensurable,
i.e., their ratio cannot be expressed as a rational fraction), the oscil
lation q is no longer periodic, since the sum of two periodic functions
with incommensurable periods is not periodic.
Expressing energy in normal coordinates. From the form of the
Lagrangian (7.29) it can be immediately concluded that the expression
for energy in normal coordinates reduces to the form
h <*>?#?), (7.31)
since L = T U and (^ = T+U. This result is true for small oscil
lations with any number of degrees of freedom.
We must note that if the normal coordinates are expressed directly
by equations (7.30), then the separate energy terms ^ (Qf + ^fQi),
will also be multiplied by certain numbers a,. However, if we replace
Qi by Qi Vo7, then these numbers are eliminated from the expression
for energy, which is then reduced to the form (7.31). An example
of this procedure is given in the exercises.
Thus, the energy of any system performing small oscillations is
reduced to the sum of the energies of separate, independent linear
harmonic oscillators. As a result of this, consideration of oscillation
problems is greatly simplified since the linear harmonic oscillator is,
in many respects, one of the most simple mechanical systems.
The reduction to normal coordinates turns out to be a very fruitful
method in studies of the oscillations of polyatomic molecules, in
the theory of crystals, and in electrodynamics. In addition, normal
coordinates are useful in technical applications of oscillation theory.
The case ol equal frequencies. If the roots of equation (7.24) coincide,
the general solution must not be written in the form (7.26), but
somewhat differently, namely,
q l = A cos co 4B sin co, )
At TV  \ ( 7  32 )
q 2 =A cos co + B sin co . J
50060
66 MECHANICS [Part I
Four arbitrary constants appear in this solution, and this is as
it should be in a system with two degrees of freedom.
An example of such a system is a pendulum suspended by a string
instead of a hinge. In the approximation (7.32), it turns out that
the pendulum describes an ellipse centred about the equilibrium
position. Account taken of the subsequent terms in the expansion
of the potential energy in powers of deflection shows that the axes
of the ellipse do not remain stationary, but rotate.
Exercise
Find the natural frequencies and normal oscillations of a double pendulum,
taking the ratios of load masses y. = 3/4 and the rod lengths X = 5/7.
From the equation for the oscillation frequencies of a double pendulum,
we obtain <o* = ~ ~r <a = T^ ~r Further, ^  7/3, 2 = 7/5.
L I 1U I
Let us now write down the expression for kinetic energy. For simplicity,
we write l = g~ ml so that only ratios of X and JJL will appear in all the equa
tions. This gives oc n 1 f (Jt 7/4, a 12 = (xX 15/28, a 22 = jiX 2 = 75/196; p n 1 +jjt =
= 7/4, p 12 = 0, p 22 = (jiX 15/28. Let us determine the coefficients a,. To do this
we must calculate the kinetic energy
7 2 15
4 l 2 14 l
V5"
Consequently, wo must put a t = ~ , a 2 = 1/2 .
Denoting   and ^ again by the letters <?i and Q 2 , we have the
A i
expression for potential energy
as it should be according to (7.10). The generalized coordinates are related to
the normal coordinates by
5\/3"/7
Thus, if 7 9 = 3 <];, and 7 9 = 3 \ initially, then we have Q 2 = for all
time, so that both pendulums oscillate with one frequency co lf with the constant
relationship between the deflection angles 7 9 3 ^ holding all the time.
Both pendulums are deflected to opposite sides of the vertical. The other normal
oscillation, with frequency w 2 , occurs for a constant angular relationship
7 9 = 5 4>.
Sec. 8. Rotating Coordinate Systems. Inertial Forces
The equivalence of inertia! coordinate systems. The particular
significance of inertial coordinate systems in mechanics was pointed
out in Sec. 2. In such systems, all accelerations are produced by
Sec. 8] ROTATING COORDINATE SYSTEMS INERTIAL FORCES 67
interaction between bodies. It is impossible to find a strictly inertial
system in nature (any system is noninertial if the motions of bodies
in it are observed over a sufficiently long period of time).
In the exercise at the end of this section we shall consider the
Foucault pendulum, whose plane of oscillation rotates with a speed
depending only on the geographical latitude of its location. This
rotation cannot be explained by an interaction with the earth, because
the gravitational force cannot make the pendulum rotate from east
to west instead of from west to east. * However, if we consider several
oscillations, then the rotation of the plane is still insignificant and
can be ignored. Then it is sufficient to consider that gravity alone
is acting on the pendulum and that the coordinate system fixed
in the earth is approximately inertial over a period of several oscil
lations.
The concept of an inertial system is meaningful as an approximation
and is a very convenient idealization in mechanics. In such a co
ordinate system, the interaction forces are measured by the acceler
ations of the bodies.
Let a coordinate system be defined for which it is known that,
to the required degree of accuracy, it can be regarded as inertial.
Then another coordinate system, moving uniformly relative to it,
is also inertial within the same degree of accuracy. Indeed, if all
the accelerations in the first system are due to interaction forces
between bodies, then no additional accelerations can appear in the
second system either. Therefore, both systems are inertial. Either
of them may be considered at rest and the other moving, since motion
is always relative.
The principle of relativity. One of the basic principles of mechanics
is that all laws of motion have an identical form in all inertial co
ordinate systems, since these systems are, physically, completely
equivalent. This principle of the equivalence of all inertial systems
is known as the relativity principle, for it is connected with the
relativity of motion.
It should be noted that this in no way signifies that inertial and
noninertial coordinate systems are equivalent: in the latter, not
all the accelerations can be reduced to interaction forces, so that
there is no physical equivalence between two such systems.
Mathematically, the principle of relativity is expressed by the fact
that equations of motion for one inertial system preserve their form
after the variables have been transformed to another inertial system.
The equations for the transformation from one inertial system
to another can be obtained only on the basis of certain physical
* In this case the plane of oscillation must pass through the vertical, since,
otherwise, the pendulum would have an initial angular momentum relative
to the vertical and would describe an ellipse whose semiaxes rotate (see end
of Sec. 7).
68 MECHANICS [Part I
assumptions. In Newtonian mechanics it is always taken that the
interaction forces between bodies, in particular, gravitational forces,
are transmitted instantaneously over any distance. Thus, the dis
placement of any body immediately transmits a certain momentum
to any other body, no matter where it is located. As a result, a clock
located in a certain inertial system can be instantaneously synchronized
with a clock moving in another inertial system. Thus, in Newtonian
mechanics, time is considered universal. In transforming from one
inertial system to another (the latter with a velocity F relative to
the former) it is taken that the time t is the same in both systems.
Later on we shall see that this assump
tion is approximate and holds only when
the relative velocity of the systems is
considerably less than the velocity of
light.
*i The Galilean transformation. Let us
I , construct coordinate systems in two
j x inertial frames of reference such that
their abscissae are in the direction of
Fig. 11 the relative velocity V and the other
coordinate axes are also mutually parallel.
Then from Fig. 11 it will be immediately seen that the abscissa of
point x in the system which we shall call stationary is related to the
abscissa in the moving system by the simple relation
x=x' + Vt, (8.1)
provided that the origins coincided at the instant t Q. The co
ordinate construction does not impose any limitations on the generality
of the transformation equations. The remaining transformations lead
simply to the identities
y = y', z=z'. (8.2)
The relationship t t' is a hypothesis which is correct only
for values of V considerably less than the velocity of light
(Sec. 20).
Condition (8.1) is absolutely symmetrical with respect to both
inertial systems: if we consider that the one in which the variables
are primed is stationary and the other in motion, (8.1) retains the
same form; one should, of course, replace F by F. In the given
case, symmetry exists because t' t. If t^t', the transformation
equations x x' + Vt and x' ~x Vt' would contradict each other.
But it would seem that equation (8.1) is obtained, quite obviously,
from Fig. 11. Thus, if we do not consider time as identical for all
inertial systems, the mathematical formulation of the relativity
principle should be more complicated than that obtained on the
basis of equation (8.1); and, we must definitely give up this "obvious
Sec. 8] ROTATING COORDINATE SYSTEMS INERTIAL FORCES 69
ness," which is so rooted in our everyday experience with velocities
that are small compared with the velocity of light.
The equations of Newtonian mechanics involve, on the righthand
side, the forces of interaction between particles. These forces depend
on the relative coordinates of the particles and, for this reason,
they do not change 'with transformation (8.1), since Vt is cancelled
in the formation of differences between the coordinates of any pair
of particles. The lefthand sides of the equations contain accelerations,
i.e., the second derivatives of the coordinates with respect to time.
But since time enters linearly in (8.1) and is the same in both systems,
x=x'. Thus, the equations of mechanics are of iden
tical form in any inertial frame of reference.
To summarize, the equations of mechanics do not
change their form when the variables undergo trans
formations (8.1). In other words, it is common to say
that the equations of mechanics are invariant to these
transformations, which are usually called Galilean
transformations .
The constancy (invariance) of mechanical laws
under Galilean transformations is the essence of the
relativity principle of Newtonian mechanics.
Here we must bear in mind that the relativity
principle, which expresses the equivalence of all iner
tial coordinate systems, expresses a far more general
law of nature than the approximate equations of transformation
(8.1), (8.2). The extension of the relativity principle to electromag
netic phenomena involves the replacement of these equations by
more general ones, which reduce to the former equations only when
all velocities are much smaller than that of light.
Rotating coordinate systems. Several new terms appear in the
equations of mechanics when transforming to rotating coordinate
systems. Let us first obtain the equations for this transformation.
In Fig. 12 the axis of rotation is represented by a vertical line.
The origin is on the axis of rotation. Let r be the radius vector
of a point A rotating around the axis. Then, for a rotation angular
velocity co (radians per second) the linear speed of the point will be
v = co r sin a , (8.3)
since the radius of rotation is p r sin a (see Fig. 12). Let the rotation
be anticlockwise. If point A lies in the plane of the paper, then the
velocity v is perpendicular to the plane of the paper and directed
towards the back of the paper. This permits us to obtain a relationship
between the linear and angular velocities in vector form. We represent
the angular velocity by a vector directed along the axis of rotation
and associated with the direction of rotation by the corkscrew rule.
70 MECHANICS [Part I
Then, if the rotation occurs in an anticlockwise direction, the vector to
is directed upwards from the paper. From this it follows that
v=[cor]. (8.4)
This expression ensures a correct magnitude and direction for the
linear velocity of the point.
Let us assume that point A, in addition to rotation, is somehow
displaced relative to the origin with velocity v' = r. The resultant
velocity of the point relative to a noiirotating system will be
represented as the sum v' + v. The kinetic energy of the point relative
to the nonrotating system is  (v + v') 2 , and the Lagrangian is
=  (v + v') 2 C7(r) = ^(v' + [cor])2J7(r). (8.5)
Let us now write down Lagrange's equations for motion relative
to a rotating system, i.e., considering r a generalized coordinate.
In order to do this we must calculate the derivatives ^ and :
or or
let it be noted that differentiation with respect to a vector denotes
a shortened way of writing down the differentiation with respect
to all of its three components. The general rules for such differentiations
will be given in Sec. 11; here we shall calculate the derivatives for
each component separately.
Let <o be along the direction of the zaxis. Then, in vector com
ponents, L will be of the form
L=  [(icot,) 2 + (y + coo:) 2 + z 2 ] U(x,y 9 z) . (8.6)
Whence we obtain
dL .. v dL /. , v dL
dL / , v 8U dL .. . dU
= m<*(y+<*x) ^, = mco (x coy) ^
dL __ dU
dz ~~ dz *
Lagrange's equations in component form appear thus:
f\ T^
7/1(0; <oi/) mco (y + cox) may + g = ,
o 7"7
m (y + coi) + mo (x co?/) + m&x + j = ,
Sec. 8] KOTATING COORDINATE SYSTEMS INERTIAL FORCES 71
Let us leave on the left only the second derivatives and rewrite
the last three equations as a single vector equation:
m[co[r<o]] . (8.7)
Expanding the double vector product on the right by means of
the equation [A [BC]] = B (AC) C (AB), and transforming to com
ponents, we can see that (8.7) is equivalent to the preceding system
of three equations. A direct differentiation with respect to the vectors r
and f would have led to (8.7), without the expression in terms of
components.
Inertial forces. The first three terms on the right in (8.7) essentially
distinguish the equations of motion, written relative to a rotating
coordinate system, from the equations written relative to a non
rotating system.
The use of a noninertial system is determined by the nature of
the problem. For example, if the motion of terrestrial bodies is being
studied, it is natural to choose the earth as the coordinate system,
and not some other system related to the Galaxy (the aggregate
of stars in the Milky Way). If we consider the reaction of a passenger
to a train that suddenly stops, we must take the train as frame of
reference and not the station platform. When the train is braked
sharply, the passenger continues to move forwards "inertially" or,
as we have agreed to say, he continues to move uniformly relative
to an inertial system attached to the earth. Thus, relative to the
carriage, it is the familiar jerk forward. At the same time it is obvious
that the noninertial system is the train and not the earth, since
no one experiencies any jerk on the platform.
The additional terms on the right of equation (8.7) have the same
origin as the jerk when the train stopped; they are produced by
noninertiality (in the given case, rotation) of the coordinate system.
Naturally, the acceleration of a point caused by noninertiality of
the system is absolutely real, relative to that system, in spite of the
fact that there are other, inertial, systems relative to which this
acceleration does not exist. In equation (8.7) this acceleration is
written as if it were due to some additional forces. These forces
are usually called inertial forces. In so far as the acceleration associated
with them is in every way real, the discussion (which sometimes
arises) about the reality of inertial forces themselves must be con
sidered as aimless. It is only possible to talk about the difference
between the forces of inertia and the forces of interaction between
bodies.
But if we consider the force of Newtonian attraction, we cannot
ignore the striking fact that, like the forces of inertia, it is proportional
to the mass of the body. As a result of this, the equations of mechanics
can be formulated in such a way that the difference between gravi
72 MECHANICS [Part I
tational forces and incrtial forces does not at all appear in the equations ;
all these forces turn out to be physically equivalent. However, this
formulation is, of course, connected with a reevaluation and a
substantial revision of the basis of mechanics. It is the subject of
Einstein's general theory of relativity, which is discussed in somewhat
more detail at the end of Sec. 20.
Coriolis force. Let us now consider in more detail the inertial forces
appearing in (8.7), which are due to a rotating coordinate system.
The first term in (8.7) occurs as a result of nonconstancy of angular
velocity. It will not interest us. The second term is called the Coriolis
force. For a Coriolis force to appear, the velocity of a point relative
to a rotating coordinate system must have a projection, other than
zero, on a plane perpendicular to the axis of rotation. This velocity
projection can, in turn, be separated into two components: one,
perpendicular to the radius drawn from the axis of rotation to the
moving point, and the other, directed along the radius. The most
interesting, as to its action, is the component of the Coriolis force
due to the radial component of velocity. It is perpendicular both
to the radius and to the axis of rotation. If a body moves perpen
dicularly to a radius, then its Coriolis acceleration is radial, and
therefore analogous in its action to the centripetal acceleration which
will be considered a little further on.
We note that the Coriolis force cannot be related, even formally,
to the gradient of a potential function U.
There are many examples of the deflecting action of the Coriolis
force in nature. The water of rivers in the Northern Hemisphere
which flow in the direction of the meridian, i.e., from north to south,
or from south to north, experience a deflection towards the right
hand bank (if we are looking in the direction of flow). This is why
the righthand bank of such rivers is steeper than the left. It is easy
to form the corresponding component of the Coriolis force. The
angularvelocity vector of the earth's rotation is directed along the
earth's axis, "upwards" from the north pole. The waters of a river,
which flows southwards at the mean latitudes of the Northern Hemi
sphere, have a velocity component perpendicular to the earth's
axis and directed away from the axis. This means that the Coriolis
acceleration of the water, relative to the earth, is in a westerly direction
or, relative to a river flowing southwards, to the right. If the river
flows in a northerly direction, the deflection will be towards the
east, i.e., again to the right. In the southern hemisphere the deflection
occurs leftwards.
The warm Gulf Stream which flows northwards is deflected towards
the east, which is of tremendous importance for the climate of Europe.
In general, the Coriolis force considerably affects the motion of air
and water masses on the earth, though when compared in magnitude
with the gravitational force it is very insignificant. Indeed, the angular
SeC. 9] THE DYNAMICS OF A RIGID BODY 73
velocity of the earth, as it completes one rotation about its axis
in 24 hours, is a little less than 10~ 4 rad/sec, while the velocity of
a particle of water or air can be taken as having an order of magnitude
of 10 2 cm/sec. From this the Coriolis acceleration has an order of
magnitude of 10~ 2 cm/sec 2 , which is one hundred thousand times
less than the acceleration caused by the force of gravity.
The Coriolis force also causes the rotation of the plane of oscillation
of a Foucault pendulum. With the aid of the Foucault pendulum,
we can prove the rotation of the earth about its axis without astro
nomical observations. In a nonrotating system, the plane of oscillation
must be invariable in accordance with the law of conservation of
angular momentum.
Centrifugal force. The third vector term in equation (8.7) is the
usual centrifugal force. Indeed, it is perpendicular to the axis of
rotation and, in absolute value, is equal to
 m [d/[tor]]  = mto  [cor]  mi* (cor sin a) = mco 2 r sin a . (8.8)
Here, the first equality takes account of the fact that the vectors to
and [tor] are perpendicular to each other, so that the absolute value
of the vector product is equal to the product of their absolute values.
But r sin a is equal to the distance from the axis of rotation, so
that this force satisfies the usual definition of a centrifugal force.
Exercise
Let us consider the rotation of the plane of oscillation of a Foucault pendulum
under the action of the earth's rotation about its axis.
The axis Ox at a given point on the earth is drawn in a northerly direction
and the axis Oy in an easterly direction. Then, if <* B = <*> sin 0, where is the
latitude of the locality, we have the equation of motion
x= 6) 2#_2cD B , y = _ .coj?/ + 2to B , co2 = J7
Multiplying the first equation by y and the second by x and then sub
tracting, we got
*(y x xy) =  A(yi+ si),^ .
Integrating and transforming to polar coordinates (x = r cos 9, y ~ r sin 9) :
r*v=r*<* B .
Whence, after cancelling the r 2 's, we have
9 = W B = <o sin ,
which gives the angular velocity of rotation of the piano of oscillation.
Sec. 9. The Dynamics of a Rigid Body
The dynamics of a rigid body is a large independent chapter of
mechanics and is very rich in technical applications. Our aim is
to give only a brief account of the basic concepts of this branch
74 MECHANICS [Part I
of mechanics inasmuch as it contains instructive examples of general
laws. In addition, certain mechanical quantities that characterize
a rigid body are necessary for an understanding of molecular spectra.
The kinetic energy of a rigid body. As was shown in Sec. 1, a rigid
body has six degrees of freedom. Three of them relate to the trans
lational motion of the centre of mass of a body in space. The re
maining three degrees of freedom correspond to rotation (relative
to this centre of mass).
In Sec. 4, it was shown that the kinetic energy of a system consists
of the kinetic energy of the motion of the whole mass of the body
concentrated at the centre of mass, and the kinetic energy of the
relative motion of the separate particles of the system. In the case
of a rigid body, relative motion reduces to rotation with the value
of angular velocity co the same for all particles. Naturally, both the
magnitude and the direction of to may vary with time.
Let us calculate the kinetic energy of rotation of a rigid body.
In the general case, the density p of the body may not be uniform
over the whole volume of the body, and may depend on the co
ordinates: p p (#, y, z) = p (r). The mass of an element of volume dV
is equal to dm~p dV. The velocity of rotation v is, from (8.4), [cor],
Therefore, the kinetic energy of the volume element is equal to
y p [cor] 2 dF. The kinetic energy of the whole body is represented by
the integral of this quantity with respect to the volume
(9.1)
Expressing the square of the vector product in terms of the compo
nents co, we have
[cor] 2 = co 2 r 2 sin 2 a = co 2 r 2 co 2 r 2 cos 2 a = co 2 r 2 (cor) 2 .
Here a is the angle between co and r. But
(cor) 2 = (cojctf + coy/ + co*z) 2 =
Since the body is rigid, the components co*, co y , co* can be taken
out of the volume integral. Combining terms which are similar in the
components co, we obtain for T:
9 xzdV
co y co* J 9 yz dV . (9.2)
Sec. 9] THE DYNAMICS OF A RIGID BODY 75
Moments of inertia. All the integrals appearing in (9.2) depend only
on the shape of the body and its density distribution, and do not de
pend on the motion of the body (in a coordinate system fixed in the
body). We denote them as follows:
=/ p (y z + * 2 ) dV, J xr =j pxydV,
(9.3)
The quantities with the same indexes are called moments of inertia,
while those with different indexes are called products of inertia.
In the notation of (9.3), the kinetic energy has the form
T = (JxxCOx + /yyCOy + JzztoZ + 2J xy 6>xCOy + 2J xz to x to z + 2j y ^CO y CO^) .
(9.4)
With the aid of the summation convention used in Sec. 2, when eval
uating Lagrange's equations the kinetic energy can be written in the
following concise form :
T^^J <**<*.
Principal axes of inertia. Let us suppose that Oxyz is a coordinate
system fixed in a body. In this system all the quantities J xx , . . . , J yz
are constant. Let us take another coordinate system Ox'y'z' which is
also fixed in the body. The old coordinates of any point are expressed
in terms of its new coordinates by the wellknown formulae of analyti
cal geometry:
x = x' cos /. (#', x) + y' cos /_ (y r , x) + z' cos /_ (z', x) ,
y = x' cos L (#', y) + y' cos /_ (y f , y) + z' cos L (*', y) ,
z = x' cos L (x', z) + y' cos /. (y 1 , z) + z' cos /_ (z f , z) ,
or, if we denote cos < (xa'x$) by the symbol A^*, then, with the aid of
the summation convention
*^3 == *^<x "oc3*
The same formulae are used to express also the components of any
vector, and in particular 03, relative to the old axes, in terms of the
components co a ' relative to the new axes.
Let us substitute these expressions into the kinetic energy (9.4)
and collect the terms containing the products eoj/fcV, co^'co/, toy'ci)/ and
76 MECHANICS [Part I
the squares to*' 2 , o> y ' 2 , to/ 2 . We shall now show that we can always
rotate the coordinate axes so that the coefficients of the new products
tox'coy', cox'co*', aVco/ become zero. Indeed, any rotation of the coordinate
system can be described with the aid of three independent parameters,
for a coordinate system is like an imaginary rigid body and its posi
tion in space is defined by the three angles of rotation (see Sec. 1).
These three angles can be chosen so that the sums of the products of
the cosines of the angles between the axes, for tox'coy', co^'ox*' and coy'co*'
become zero. The remaining expressions for co*' 2 , co y ' 2 , and to*' 2 will be
called J l9 J 2 , / 3 , so that
The kinetic energy is written in the following form in the new coordi
nate axes:
T  \ (J>? + J*<*\ + Js<t). (0.5)
These axes are called the principal axes of inertia of the body, they
can be defined relative to any point connected with the body. By defi
nition, the products of inertia convert to zero in the principal axes of
inertia. The moments of inertia in the principal axes are called princi
pal moments of inertia. They are denoted by J l9 J 2 , J 3 .
The angular momentum of a rigid body. Let us now calculate a pro
jection of the angular momentum of a rigid body. From the definition
of angular momentum we obtain
M* = J p [rvMF = J p [r [tor]] x dF  J p (to. v r 2 x (cor)) dV =
^ cox Jp(2/ 2 + z 2 )dV co y f pxydV <o* f pxzdF =
= Jxx CO* + + J xy C0 y + J xz 6)z (9.6)
or, in shortened form,
Comparing (9.6) and (9.4), we see that
^T 1
*;'
My and M z appear analogous. In vector form, we may write
(9.8)
Equations (9.7) and (9.8) again express the fact that the angular
momentum is a generalized momentum related to rotation. In this
sense, (9.7) corresponds to (5.4). The only difference is that the com
ponents to are not total time derivatives of some quantities. This will
Sec. 9] THE DYNAMICS OF A RIGID BODY 77
be shown a little later in the present section. In that sense, to*, in
(9.7), is not altogether similar to 9 in (5.4).
If the coordinate axes coincide with the principal axes of inertia,
then the expression for angular momentum is even simpler than
(9.6):
AT
*i = ls: = Jii (9.9)
and similarly for the other components.
Moment of forces. Let us now find equations which describe the
variation of angular momentum with time. The derivative of angular
momentum of a particle is
where the first term becomes zero since r and p are parallel. Integrating
this equation over the volume of the rigid body and taking advantage
of the additive property of angular momentum, we have
K. (9.10)
The righthand side of (9.10), which we denote by K, is called the
resultant moment of the forces applied to the body. If F is the gravi
tational force (which occurs in the majority of cases) then K can also
be written as
= Jp<7[rz ]dF,
where z is the unit vector in a vertical direction. But since the vector
z is a constant, it should be put outside the integration sign:
K = [z , JpgrrdFJ,
If the body is supported at its centre of mass, then, by the definition
of centre of mass, the integral for all three projections pr will be zero.
Then K = and the total angular momentum will be conserved. This
occurs in the case of a gyroscope.
For the conservation of angular momentum of a rigid body it is
sufficient that K = 0; but for any arbitrary mechanical system,
angular momentum is conserved only when there are no external
forces.
Euler's equations. Equation (9.6) gives a relationship between M
and co. The quantities /**, . . ., J yz are constant only in a coordinate
system fixed in the rigid body itself. If we write equation (9.10) for
a stationary coordinate system, then, differentiating M with respect
to time, we must also find the derivatives of J xx , . . . , J YZ with respect
to time, which is very inconvenient. Therefore, it is preferable to
78 MECHANICS [Part I
transform the equation to a coordinate system fixed in the body,
taking into account the accelerated motion of that system. The varia
tion of the vector M relative to the moving axes consists of two com
ponents : one is due to the variation of the vector itself, while the other
is due to the motion of the axes onto which it is projected. For the vector
M this variation is equal to [u>M], similar to the way that it was equal
to [cor] for the radius vector r in Sec. 8. When the coordinate system
is rotated, any vector varies like a radius vector.
Let the coordinate axes be taken in the direction of the principal
axes of inertia. Obviously, the moments of inertia relative to these
coordinates are constant. For this reason, the time derivative of
M l = J l <f) 1 is
M l = J c^ + [coM]! = J 1 v l + o> 2 M 3 co 3 M 2 = Jjco,, + (J 3 J 2 ) co 3 o> 2 .
Equating this expression to the magnitude of the projection of the
moment of force on the first axis of inertia, and doing the same for
the other axes, we obtain the required system of equations
(9.11)
These equations were obtained by L. Euler and are named after
him. They can be reduced to quadrature for any arbitrary values of
integrals of motion in the following cases :
1) JK 1 K 2 ~K 3 = (point of support at the centre of mass) for
arbitrary values of the moments of inertia;
2) J 2 = J 3 ^ Jj and the point of support lies on the axis of symmetry,
relative to which two moments of inertia are equal. This is the so
called symmetrical top.
For more than a hundred years, no other case of a solution of system
(9.11) by quadratures was known. Only in 1887 did S. V. Kovalevskaya
find another example (see G. K. Suslov, Theoretical Mechanics,
Gostekhizdat, 1944). Kovalevskaya showed that the three listed cases
exhaust all the possibilities of integrating the system (9.11) by quadra
tures for arbitrary constants (integrals) of motion.
A free symmetrical top. All three cases, and in particular the Kova
levskaya case, are very complicated to integrate. Therefore, we shall
only consider the simplified first case, when J 2 = J 3 (a free symmetrical
top).
From the first equation of (9.11), it immediately follows that
a> 1= = const. For brevity, we write the value
Q. (9.12)
The second two equations of (9.11) are written thus:
Sec. 9]
THE DYNAMICS OF A RIGID BODY
79
<J> 2 4 }<o 3 = , o> 3 ico 2 = 0. (9.13)
Equations (9.13) are easily integrated if we represent the components
co 2 and co 3 in the following form:
= coj^ cos fit , co 3
(9.14)
Here, o> + co = co^ is a constant quantity. Thus, the angular
momentum projection on the axis of symmetry and the sum of the
squares of the angularmomentum projections on the other two axes
are conserved. This means that the angularmomentum vector rotates
about the axis of symmetry, i. e., the first axis of inertia, with angular
velocity fl; the vector makes with it a constant angle, the tangent
of which is ^~ . This is the situation in a system of moving axes.
Of course, in a system of stationary axes, the total angular momen
tum is conserved in magnitude and direction, since the resultant
moment of force is equal to zero. In this
system, the axis of symmetry of the top
rotates about the angularmomentum
direction making a constant angle with
it. Such motion is called precession. Pre
cessional motion is only stable for rela
tively small external perturbations. The
stabilizing action of gyroscopes is based
on this principle.
Eulerian angles. We shall now show
how to describe the rotation of a rigid
body with the aid of parameters which
specify its position. Such parameters are
the Eulerian angles shown in Fig. 13. The
figure depicts two coordinate systems:
a fixed system Oxyz and a system
Ox'y'z' fixed in the rigid body. It is most convenient to take x', y', z'
along the principal axes of inertia through the point of support. Then
the Eulerian angles are:
& is the angle between the axes z and z',
9 is the angle between the line OK of intersection between the planes
xOy and x'Oy' and the #'axis,
fy is the angle between the line OK and the #axis.
If the angle fy varies, then the angular velocity vector <J/ is directed
along the axis Oz since that vector is perpendicular to the plane of
angle of rotation ^. Thus 9 must be taken along the axis Oz' and &
along the line OK.
Let us now express the angularvelocity projections (i. e., & 19 co 2 ,
co 3 ,) onto the principal axes of inertia in terms of the generalized ve
locities 4, <p, &
80 MECHANICS [Part I
co 3 is the projection of the angular velocity on the axis Oz f (z r is
the third axis). As was shown, 9 is projected exclusively on this axis
and the projection of fy is equal to ^ cos &, since # is the angle between
the axes Oz and Oz r . Hence,
0)39 f^cosfl . (9.15)
In order to find the projections of the angular velocity on the other
two axes, we draw a line OL which lies in the plane x'Oy' and is per
pendicular to OK.
From Fig. 13 it can be seen that
/_LOx' = J 9 and /_zOL = ^ + a,
since the straight line OL lies in the plane zz' ', as do all lines perpendic
ular to OK. The projection of fy on OL is equal to A sin &, and the
projection on Ox' is equal to ^ sin ^ cos iy 9)= <j> sin &
sin 9. The projection of ^ on Oy' is ^ sin 9 cos 9. The projection of &
on Ox' and Oy' can be directly found by means of the diagram ; they
are cos 9 and 9 sin 9. The result is therefore
co! & cos 9 ^sin#sin9 , (9.16)
(9.17)
From equations (9.15), (9.16), and (9.17) it will be seen that co 1?
co 2 and co 3 are not total time derivatives of any quantities and, in
that sense, do not exactly agree with the usual notion of generalized
velocities (as do 9, ij;, $).
If we substitute into (9.5) the expressions for co 1 , o> 2 , co 3 in terms of
the Eulerian angles, we obtain the kinetic energy of a rigid body as a
function of the generalized coordinates 9, <p, &.
The symmetrical top in a gravitational field. We shall find the Lagran
gian for a symmetrical top whose point of support lies on the axis of
symmetry at a distance I below the centre of mass. Then the height
of the centre of mass above the point of support is z = l cos #. Hence,
the potential energy of the top is
U = mgz = mgl cos 0 . (9.18)
The kinetic energy of the top, expressed in terms of the Eulerian
angles, is
T = ~ J t (co? + col) + 4^3^1 =
M
= yM^ 2 + <Psin2) + ^ J 3 (^ + <j,cos) 2 . (9.19)
The difference between the quantities (9.19) and (9.18) gives the
Lagrangian for a symmetrical top. The sum gives the total energy & .
Sec. 10] GENERAL PBINCIPUSS OF MECHANICS 81
Since L does not contain time explicit^, the energy is an integral
of motion:
= T + U = const . (9.20)
We can find two more integrals of motion, noting that the angles 9
and <J> do not appear explicitly in L (9 is eliminated only in the case
of a symmetrical top). These integrals of motion are
p, 7"
#p = p = ^3 (? + ^ cos *) = const , (921)
o i"
p^ =  = e/x sin 2 ^^  J 3 cos ft (9 + <j> cos ft) = const . (9.22)
If we eliminate 9 and ^ from equations (9.21) and (9.22) and substi
tute them into the energy integral, the latter will contain only the
variable #, which allows us to reduce the problem to quadrature.
Substituting (9.21) in (9.22), we obtain
p^ = J l sin 2 & fy f fa cos & ,
whence
The energy integral, after substituting p 9 and p^ is
**"** + * + mgl cos . (9.23)
Thus, the problem is reduced to motion with one degree of freedom
ft, as it were. The corresponding "kinetic energy" is J^ 2 , and the
' 'potential energy" is represented by those energy terms which depend
on &. This potential energy becomes infinite for = 0, and & n.
Hence, for <0 <TT it has at least one minimum. If this minimum
corresponds to 9 > > then the rotation of the top, whose centre of
mass is above the point of support, is stable. Small oscillations are
possible near the potential energy minimum. These oscillations are
superimposed on the precessional motion of the top which we have
already noted. They are called nutations.
Sec. 10. General Principles of Mechanics
In this part of the book, mechanics is explained mainly through the
use of Newton's equations (2.1). Going over to generalized coordinates,
we obtain from them Lagrange's equations and a series of further
deductions. In this section it will be shown that the system of Lagran
ge's equations can be obtained not only from Newton's Second Law,
but also from a very simple assertion about the value of the integral
60060
82 MECHANICS [Part I
of the Lagrangian taken with respect to time. The basic laws of mechan
ics thus formulated are usually called integral principles.
The particular importance of these principles is that they allow
us to understand, in a unified manner, the laws relating to various
areas of theoretical physics (mechanics and electrodynamics), thus
opening up a field for broad generalizations.
Action. For a certain mechanical system, let it be possible to define
the Lagrangian
L=L(q,qJk), (10.1)
as dependent on the generalized coordinates q, velocities q, and the
time t. We shall consider that all the coordinates and all the velocities
are independent. Let us choose some continuous, but otherwise arbi
trary, dependence of the coordinates upon the time q (t). The functions
q (t) can be in complete disagreement with the actual law of motion.
The only requirement imposed on q (t) is that the functions q (t) should
be smooth, i. e., that they should provide for differentiation and should
correspond to the rigid constraints present in the system.
The time integral of the Lagrangian is called the action of the sys
tem:
'i
S = JL(q, q, t) dt. (10.2)
'o
The magnitude of this integral depends upon the law chosen for
q (t), and is, in that sense, arbitrary. In order to examine the relation
ship between the action and the function q (t), it is convenient to cal
culate the change of S for a transition from some arbitrary law q (t)
to another, infinitely close but also arbitrary, law q f (t).
Variation. Fig. 14 shows two such conceivable
paths. Time is taken along the abscissa, and
one of the generalized coordinates q, represent
ing the totality of generalized coordinates, is
plotted on the ordinate axis.
For the specification of future operations, we
shall consider that both paths pass through the
same points, g and q l9 at the initial and final
instants of time,
p. 14 The vertical arrow shows the difference
between two conceivable, infinitely close paths at
some instant of time other than initial or final.
This difference is usually called the variation of q and is denoted by
8q. The symbol S should emphasize the difference between variation
and the differential d\ the differential is taken for the same path at
various instants of time, while the variation is taken for the same
instant of time between different paths.
Sec. 10] GENERAL PRINCIPLES OF MECHANICS 83
Since the neighbouring paths in Fig. 14 have different forms, the
speed of motion along them will also differ. Together with the variation
of the coordinate 8q between paths, we can also find the variation
in velocity 8q. We shall show that 8q = jr8q. Indeed, 8q = q' (t) q (t),
at
where q' and q are values of the coordinates for neighbouring paths.
But the derivative of the difference = q is equal to the difference of
at * ^
the derivatives q' (t) q(t)=8q.
Let us now find the variation of the Lagrangian, i. e., the difference
of the function for two adjacent paths. Since L = L (q, q, t) and the
variation is taken at the same instant of time, i. e., S = 0, we obtain
Let us rearrange the second term. Taking advantage of the fact that
q = 8q , we can write it thus :
The last equation simply expresses a transformation by parts.
Substituting it into (10.3), we find
d
The integral of the variation of L is equal to the variation of action
8$, since the difference between integrals taken between the same
limits is equal to the difference between the integrands.
The first term in (10.4) can be integrated with respect to time,
because it is a total derivative. The variation of action is then reduced
to the form
'o 'o
We have agreed to consider only those paths which pass through the
same points, q Q and q l9 at the initial and final instants of time. Hence,
at these instants the variation $q becomes zero by convention, and
the integrated term disappears. The expression 8$ is reduced to the
following integral:
. (10.6)
The extremal property of action. If the chosen path coincides with
the actual path of motion, the coordinates satisfy Lagrange's equation:
84 MECHANICS [Part I
Substituting this in (10.6), we see that the variation of action tends
to zero close to the actual path. The change in magnitude is equal to
zero either close to its extreme, or close to the "stationary point"
(for example, the function ?/ # 3 has such a point at x = Q, where
2/' = 0, iz/" = 0). Three cases can, in general, be realized: a minimum,
a maximum and a stationary point.
For example, let a point, not subject to the action of any forces
other than constraint reactions, move freely on. a sphere. Then its
path will be an arc of a great circle. But through any two points on
the sphere there pass two arcs of a great circle representing the largest
and smallest sections of the circumference. One corresponds to a maxi
mum, and the other, to a minimum, 8. If the beginning arid end of the
path are diametrically opposite, the result is a stationary point.
The principle o! least action. We have proven, on the basis of Lag
range's equations, that 8S 0. We can proceed in a different way:
by asserting that close to the actual path passing between the given
initial and final positions of the system the increment of action is
equal to zero, we can derive Lagrange's equations. Ordinarily, the
action on an actual path is minimal, and therefore the assertion we
have made is called the principle of least action. Action was written
in the form (10.2) by Hamilton. Much earlier, the principle of least
action was mathematically formulated by Euler for the special case
of paths corresponding to constant energy.
For us, it is not essential that the action should be a minimum, but
that it should be steady, 88 = 0.
Lagrango's equations are derived from the principle of least action
by means of proving the opposite. We assume the righthand side of
equation (10.6) to be zero, 8$ = 0, and the variation 8q to be arbitrary.
Then, if the expression inside the parentheses is not equal to zero,
the sign of the variation 8q can always be chosen to be the same as
7\ T A y\ T
for the quantity ~ ^ ^. , because the variation is arbitrary. If,
f\ T ft P 7"
for example, the sign of the quantity ^  ^ ^p changes several
times along the path of integration, then the sign of 8q must also be
changed accordingly at those points so that the integrand of (10.6)
should everywhere be nonnegative. But the integral of a nonnega
tive function cannot equal zero unless the function is equal to zero
everywhere. Therefore, 8$ = only when^ rr^ becomes zero
^ oq at oq
along the whole path of integration, for otherwise the variation 8q
can be so chosen that 88 >0. We have shown that if we proceed
from the principle of least action as a requirement for the motion along
an actual path, then that path must satisfy Lagrange's equations.
Sec. 10] GENERAL PRINCIPLES OF MECHANICS 86
The advantages of using action. The principle of least action may at
first sight appear artificial or, in any case, less obvious than Newton's
laws, to whose form we are accustomed. For this reason we shall try
to explain where its advantages lie.
First of all, let it be noted that Lagrange's or Newton's equations
are always associated with some coordinates whose choisc is, to a
significant extent, arbitrary. In addition, the choice of coordinate
system, relative to which the motion is described, is also arbitrary.
Yet the motion of particles along actual paths in a mechanical sys
tem expresses a certain set of facts which cannot depend on the
arbitrary manner of their description. For example, if the motion
leads to a collision of particles, that fact must always be represented
in any description of the system.
But it is precisely the integral principle that is especially useful in a
formulation of laws of motion not related to any definite choice of
coordinates, the value of the integral between the given limits being
independent of the choice of integration variables. The extremal
property of an integral cannot be changed by the way in which it is
calculated.
The integral principle S$ = is equivalent, purely mathematically,
to Lagrange's equations (2.21). But in order to apply it to any actual
system, we must have an explicitly expressed Lagrangian. It may be
found from those physical requirements which should be imposed on
an invariant law of motion that is independent of the choice of coordi
nate axes and the frame of reference.
As a result of the invariance of the principle of least action, we can
consider the laws of mechanics in a very general form, and this,
therefore, opens the way for further generalizations.
The determinacy of the Lagrangian. Before finding an explicit
form for the Lagrangian, we must put the qiiestion: Is the determined
function we are looking for singlevalued ? We shall show that if we
add the total time derivative of any function of coordinates and time,
IT / (q, t), then Lagrange's equations remain unchanged. This can be
verified either by simple substitution into (10.7), or directly from the
integral principle. Writing
L=L'+f(q,t), (10.8)
we see that
'i 'i 'i ti
(10.9)
The variations of / appear in the variation of S only at the limits of
integration. But since we have arranged that / depends on the coordi
86 MECHANICS [Part I
nates and time, but not on the velocities, the variation of / is expressed
linearly in terms of the variations of the coordinates, and is zero at
the limits of integration. Therefore,
(10.10)
Hence, the Lagrangian is determined only to the accuracy of the
total time derivative of the function of coordinates and time.
Defining forms of the Lagrangian. We shall now formulate in more
detail those requirements which the integral principle expressing laws
of mechanics must satisfy.
First of all we note that the form of this principle must be the same
for different inertia! systems, since all such systems are equivalent.
This statement follows from the relatively principle (see Sec. 8).
The essence of the relativity principle consists in the fact that the
choice of an inertial coordinate system is arbitrary, while the physical
consequences of the equations of motion cannot be arbitrary.
Similarly arbitrary is the choice of the origin and the initial instant
of time and also orientation of the coordinate axes in space.
It must, of course, be borne in mind that the form of action is by no
means determined by speculation ; this form represents no less a gener
alization of physical experience than the laws of Newton. However,
the principle of least action, best expresses the invariance of physical
laws to the method of their formulation. Quite naturally, the form of
the invariance (in relation to rotations, translations, reflections, etc.)
is itself a certain, very broad, generalization of experience, and must
by no means be considered as a priori.
Considering now the problem of finding the form of the Lagrangian,
let us first of all determine the action of a free particle in an inertial
coordinate system.* In such a system, the particle moves uniformly
in a straight line, i. e., with constant velocity. (This statement is based
on the experimental fact that inertial systems exist in nature). Thus,
the Lagrangian for a free particle in an inertial system cannot con
tain any coordinate derivatives other than velocity.
By definition, a free particle is very far away from any other bodies
with which it could interact. Therefore, its Lagrangian must not change
its form upon displacement of the origin to any arbitrary point fixed
in the given inertial system. In other words, the Lagrangian of such a
particle does not depend explicitly on the coordinates.
In this way, one can conclude that the Lagrangian does not depend
explicitly on time.
* See L. D. Landau and E. M. Lifshits, Mechanics, Fizmatgiz, 1958.
Sec. 10] GENERAL PRINCIPLES OF MECHANICS 87
The orientation of the coordinate axes is arbitrary as well as the
choice of the origin. For the Lagrangian to be independent of the orien
tation of coordinate axes, it must be scalar quantity.
To summarize, then, the Lagrangian is a scalar that depends only
on the velocity of the free particle relative to the given inertial system.
The only scalar quantity which can be formed from a vector is the
absolute value of the vector. Therefore,
The form of this function can be found from the relativity principle,
in accordance with which the Lagrangian must not change with the
transformation from one inertial system to another. In Newtonian
mechanics, this transformation is effected with the aid of equations
(8.1), (8.2), i.e., Galilean transformations. The Galilean transforma
tions led to the law of addition of velocities :
where V is the relative velocity of the inertial systems. Therefore,
the Lagrangian must remain invariant with respect to Galilean trans
formations.
Since the Lagrangian is determined to a total derivative, it is suffi
cient (for its invariance) for the following equality to be satisfied:
= L(^) = L[(v' + V) 2 ] = L(i;' 2 ) + g ) (J0.ll)
where the functions L (v 2 ) and L (v' 2 ) have the same form in accordance
with the principle of relativity.
Any transformation (8.1), (8.2), in which the relative velocity V
is finite, can be obtained by a set of infinitely small transformations
applied successively. It is, therefore, sufficient to consider a transforma
tion in which the relative velocity of the inertial systems V is very
much smaller than the particle velocity v. Then, to a very good approx
imation, the quantity (v'+V 2 ) is equal to
where the term of the second order of smallness is discarded.
Expanding L [(v' + V) 2 )] in a series, we obtain, to the same approx
imation,
Comparing this with (10.11), we find:
3L gy'Y dL gy<* r/
0(v' a ) "" d(v'*) V dt ~~ dt
88 MECHANICS [Part I
However, the expression on the lefthand side of the equation can
/) 7
be a total derivative of the function of coordinates only if g /a is
independent of velocity. Introducing the notation
BL m ,
7jT^r 2 y = y = const ,
we obtain
di~ ~~~di m '
for, otherwise, . ( , 2) could not be put inside the derivative sign.
In this way we have shown that the Lagrangian for a free particle
is equal to
( V 2) = ^ r a. (10.12)
The Lagrangian for a system of noninteracting particles is equal
to the sum of the Lagrangians of these particles taken independently,
since it is the only sum of quadratic expressions of the type (10.12)
that changes by a total derivative when Vi==v, / +V (where i is the
particle number) is substituted.
In order to write down L for a system of interacting particles, we
must, of course, make certain physical assumptions about the nature
of the interaction.
1) The interaction does not depend on the particle velocities. This
assumption is justified for gravitational and electrostatic forces, and
is not justified for electromagnetic forces. It should, however, be noted
that electromagnetic interactions involve ratios of particle velocities
and the velocity of light c, and therefore, to the approximation of
Newtonian mechanics, they must be considered as negligibly small.
The Lagrangian of Newtonian mechanics is not universal and is appli
cable only to a limited group of phenomena, when all V{ <^ c.
2) The interaction docs not change the masses of the particles.
3) The interaction is invariant with respect to Galilean transforma
tions.
From these conditions it can be seen that the interaction appears
in the Lagrangian in the form of a scalar function determined only
by the relative distribution of the particles:
(10.13)
From this expression, we can find the conservation laws for energy,
linear momentum, and angular momentum (see Sec. 4).
The Hamiltonian function. We shall now use the principle of least
action in order to transform a system of equations of motion to other
variables. Namely, in place of coordinates and velocities we shall
Sec. 10] GENERAL PBINCIPLES OF MECHANICS 89
employ coordinates and momenta. Let us assume that velocities are
eliminated from the relations
P = ff. (10.14)
Since the Lagraiigian depends quadratically on the velocities, equa
tions (10.14) are linear in the velocities and can always be solved. We
shall obtain for coordinates and momenta a more symmetrical system
of equations than Lagrange's equations.
The passing from velocities to momenta was performed to some ex
tent when we substituted the integrals of motion in the expression for
energy, for example, in (5.4), (9.21), (0.22).
Now, in place of the velocities we shall introduce into the energy
the momenta for all the degrees of freedom, (and not only for the cyclic
ones, i. e., those, whose coordinates do not appear explicitly in L).
Energy expressed in terms of coordinates and momenta only is called
the Hamiltonian function of the system or, for short, the Hamiltonian :
<?[q,q(p)]^Jf(q,p)=qpL. (10.15)
Thus, for example, if we replace & by ~ in (9.23), we obtain the
Hamiltonian for a symmetrical top:
(10.16)
Hamilton's equations. In order to derive the required system of
equations, we write the expression for the principle of least action,
expressing L in terms of jtf* :
ti
(10.17)
Here it is assumed that q is expressed in terms of p and q.
Let us calculate the variation 88:
The second term inside the parentheses can be integrated by parts,
similar to the way that it was done in (10.5). This gives
The integrated part becomes zero when limits of integration have been
substituted. The independent variables are now p and q. The variation
90 MECHANICS [Part I
of p, as well as the variation of q, is completely arbitrary in sign. For
88 to bo equal to zero, the following equations must be satisfied:
This system of equations is more symmetrical than Lagrange's
equations. Instead of v secondorder Lagrangian equations, we have
2 v firstorder equations (10.18). They are called Hamilton's equa
tions.
Reducing the order with the aid ol the energy integral. If #F does not
depend on time, we can exclude time completely from the equations
by dividing all the equations (10.18), except one, by the said equation.
Then we have
if w ( 10  19 >
dp
Here, for simplicity, this operation has been performed for a system
with one degree of freedom. The integration of (10.19) yields one con
stant. The second constant will be determined by quadrature from the
equation
(10.20)
where  is a certain function q which can be obtained by integrating
(10.19). The constant of integration in (10.20) is the initial instant .
The connection between momentum and action. We shall now show
that if action is calculated for the actual paths of a system, then mo
mentum can be very simply expressed in terms of this action. For this
we shall consider the change in action when the ends of the integra
tion interval are displaced along the actual paths. From (10.7), the
expression under the integral sign in (10.5) is equal to zero on such
paths. But the integrated part does not become zero ; only the varia
tions in it must be replaced by differentials, since we are considering
the displacement of the ends of the integration interval along given
paths. Therefore,
dS^^dq^dq Q =pdqp Q dq (10.21)
in agreement with the definition of momentum (4.13).
But action calculated along an actual path is uniquely determined
by its initial and final points 8 = 8 (q Q , q). So
dS = dq + 8 dq. (10.22)
SeC. 10] GENERAL PRINCIPLES OF MECHANICS 91
Comparing (10.21) and (10.22), we obtain the very important rela
tionship between momentum and action
* * (10  23)
which is very essential for the formulation of quantum mechanics.
Exercise
Write down the Hamiltonian and Hamilton's equations for a particle
in a central field.
PART II
ELECTRODYNAMICS
Sec. 11. Vector Analysis
The equations of electrodynamics gain considerably in conciseness
and vividness if they are written in vector notation. In vector notation,
the arbitrariness associated with the choice of one or another coordi
nate system disappears, and the physical content of the equations
becomes more apparent.
We have assumed that the reader is acquainted with the elements of
vector algebra, such as the definition of a vector and the various forms
of vector products. However, in electrodynamics, vector differential
operations are also used. This section is devoted to a definition of vec
tor differential operations and to proofs of their fundamental proper
ties, which will be needed later.
The vector of an area. We first of all give a definition of the vector
of an elementary area rfs. This is a vector in the direction of the normal
to the area, numerically equal to its surface and related to the
direction of traverse of the
contour around the area by
the corkscrew rule (Fig. 15).
d*
Fig. 15
Fig. 16
We shall make use of a righthanded coordinate system x, y, z, in
which, if we look from the direction of the zaxis, the araxis is rotated
towards the t/axis in an anticlockwise sense (Fig. 16). In this system,
Sec. 11] VECTOR ANALYSIS 93
the vector area can be resolved into components which are expressed
thus :
dsxdydz, ds y = dzdx, ds x =dxdy.
Vector flux. Now suppose that a liquid of density 1 ("water")
flows tlirough the area, the flow velocity being represented by the
vector v. We shall call the angle between v and ds, a. Fig. 17 shows
the flow lines of the liquid passing through ds.
They are parallel to the velocity v. Let us calcu
late the amount of liquid that passes tlirough the ds^
area ds every second. Obviously, it is equal to the
amount that passes through the area ds', placed ^
perpendicular to the flux and intersected by the
same flow lines as pass through ds. This quantity is
simply equal to v ds', because every second a
liquid cylinder of base ds' and height v passes
through the area ds'. But ds' ds cos a, whence j?ig. 17
the quantity of liquid we are concerned with is
dJ=v ds' = v ds cos a^=vds. (11.1)
By analogy, the scalar product of any vector A (taken at the point
of infinitesimal area) on ds is called the flux of the vector A across
the area ds. Similar to the way that the flow of liquid across a finite
area s is equal to the integral of dJ with respect to the surface,
JJvds, (11.2)
the integral
(11.3)
is called the flow (flux) of the vector A across any area.
The area vector is introduced so that we can make use of the
noncoordinate and convenient notation of (11.3). The integrals
appearing in (11.3) are double. In terms of the projections of (11.3)
we can write
J=\ AdsE=J \Axdydz + (JA y dzdx
where the limits of the double integrals are determined from the cor
responding projections, onto the coordinate planes, of the contour
bounding the surface.
The GaussOstrogradsky theorem. Let us now calculate the vector
flux through a closed surface. For this we shall consider, first of all,
the infinitesimal closed surface of a parallelepiped (Fig. 18). We
shall make the convention that the normal to the closed surface
will always be taken outwards from the volume.
94 ELECTRODYNAMICS [Part II
Let us calculate the flux of the
vector A across the area A BCD
(the direction of traverse being in
agreement with the direction of the
normal). Since the flux is equal to
the scalar product of A by the vec
tor area A BCD, in the negative
^direction (and hence equal to
dydz), we obtain for this infinitely
small area
Fig. 18
dJABCD = A x (x) dy dz.
We get a similar expression for the area A'B'C'D', only in this case
the projection ds x is equal to dy dz, and A x is taken at the point
x + dx instead of x. And so
dJA'ffc'D' A x (x + dx) dy dz.
Thus the resultant flux through both areas, perpendicular to the
#~axis, is
+ dJADCD = [A x (x + dx) A x (x)] dy dz = dxdydz.

(11.4)
We have utilized the fact that dx is an infinitely small quantity,
and we have expanded A x (x + dx) in a series. The resultant fluxes
across the boundaries perpendicular to the y and z axes are formed
similarly. The resultant flux across the whole parallelepiped is
A finite closed volume can be divided into small parallelepipeds,
and the relationship (11.6) applied to each one of them separately.
If we sum all the fluxes, the adjacent boundaries do not give any
contribution, since the flux emerging from one parallelepiped enters
the neighbouring one. Only the fluxes through the outer surface
of the selected volume remain, since they are not cancelled by others.
But the righthand sides of ( 1 1.6) will be additive for all the elementary
volumes dV dx dy dz, yielding the very important integral theorem:
 <" 6 >
It is called the GaussOstrogradsky theorem.
The divergence of a vector. The expression appearing on the right
hand side under the integral sign can be written down in a much
shorter form. We first of all notice that it is a scalar expression,
since there is a scalar on the lefthand side in (11.6) and dV is also
Sec. 11] VECTOR ANALYSIS 95
a scalar. This expression is called the divergence of the vector A
and is written thus:
The divergence can be defined independently of any coordinate
system, if (11.5) is used. Indeed, from (11.5) the definition for diver
gence follows as
fAds
divA=sUm^~ . (11.8)
F>0 V
The divergence of a vector at a given point is equal to the limit
of the ratio of the vector flux through the surface surrounding the
point to the volume enveloped by the surface, when the surface is
contracted into the point.
Let us suppose that the vector A denotes the velocity field of
some fluid. Then, from the definition (11.8), it can be seen that the
divergence of the vector A is a measure of the density of the sources
of the fluid, for it is obvious that the more sources there are in unit
volume, the more fluid will flow out of the closed volume. If div A
is negative, we can speak of the density of vents. But it is more
convenient to define the source density with arbitrary sign. We
note that from (11.7) there follows the quantity
since r has components x, y, z.
Contour integrals. We shall now consider the vector integral of
a closed contour having the following form:
z). (11.10)
This single integral is called the circulation of the vector over the
given contour. For example, if A is the force acting on any particle,
then A dl = A dl cos a is the work done by the force on the contour
element dl and C is the work performed in covering the whole contour.
Stokes 5 theorem. We shall now prove that the circulation of the
vector A around the contour can be replaced
by the surface integral "pulled over" the y
contour.
Let us consider the projection of an in
finitely small rectangular contour onto the
plane yz. Let this projection also have the
form of a rectangle shown in Fig. 19. We
shall calculate the circulation of A around Fig. 19
96 ELECTRODYNAMICS [Part II
this rectangle. The side A B contributes a component A y (z) dy
arid side CD the component A y (z + dz)dy, where the minus
sign must be written because the direction of the vector CD is
opposite to that of the vector AB. We obtain, for the sum due to
the sides AB and CD,
(wo have expanded A y (z + dz) in a series for dz), while for the sides
BC and DA,
dA
A; (y + dy) dz A z (y) dz   Q ? dydz .
The resultant value for circulation in the yzplane is
The notation B x is clear from the equation. Let us now find out
what meaning this expression has. From the definition of (11.10),
circulation is a scalar quantity and, hence, on the right side of equation
(11.11) there must also be a scalar quantity. If the contour lies in
the plane yz, this quantity is of the form dC~B x ds x \ consequently,
for an arbitrary orientation of the contour, the relationship (11.11)
must have the form of the scalar product
where B x , B y , B z must necessarily be the components of a vector,
since, otherwise, dC could not be a scalar. From (11.11),
/? 3A Z _ dA y . ~.
*~~~~W. aT (ii.id)
In order to find r the circulation for infinitely small contours in
the xz, yz planes, it is sufficient to perform a cyclic permutation
of the indices x, y, z. This permutation yields the components B y , B z :
The vector B has a special name: it is called the rotation or curl
of the vector A and is denoted thus:
B=rot A.
rot A is expressed in terms of unit vectors i, j, k, directed along
the coordinate axes:
C. 11] VECTOR ANALYSIS 97
Changing to the notation (11.16), we see that the component of
rot A normal to the area appears in equation (11.11):
J
A dl = rot* A <fa , (11.17)
where the subscript n of rot A indicates that we must take the pro
jection of rot A normal to the area, i.e., coinciding with the vector ds.
(11.17) permits us to define rot A in a noncoordinate manner, similar
to the way that we defined div A in (11.8), namely:
(Adi
rot n A = lim   , (11.18)
5 > ' 9
or the projection of rot A, normal to the area at the given point,
is the limit of the ratio of the circulation of A, over the contour
of the area, to its value when the contour is contracted into the point.
So that the integral JA dl should not become zero, we must have
closed vector lines, to some extent following the integration contour,
which lines are simila.r to the closed lines of flow in a liquid during
vortox motion. Hence the term curl, or rotation.
If the circulation is calculated from a finite contour then the contour
can be broken up into infinitely small cells to form a grid. For the
sides of adjacent cells, the circulations mutually cancel since each
side is traversed twice in opposite directions; only the circulation
along the external contour itself remains. The integral on the right
hand side of equation (11.17) gives the flux of rot A across the surface
"pulled over" the contour. Thus, we obtain the desired integral
theorem
[AdlJrotAds, (11.19)
which is called Stokes' theorem.
Differentiation along a radius vector. The divergence and rotation
of a vector are its derivatives with respect to the vector argument.
They can be reduced to a unified notation by means of the following.
We introduce the vector symbol V (nabla*) with components
Then, from (11.7), we obtain for the divergence of A:
* Nabla is an ancient musical instrument of triangular shape. This symbol
is also called del.
70060
98 ELECTRODYNAMICS [Part II
(11.21)
x 
i.e., a scalar product of nabla and A.
From (11.16), we have for the rotation
rot A ~ i (Vy A x V z A y ) t j ( V* A x V* A z ) + k ( V* A Y Vy A x
LVA]. (11.22)
We use the identity symbol . here in order to emphasize the fact
that we are simply dealing with a new system of notation. We shall
see, however, that this system is very convenient in vector analysis.
We note, with reference to algebraic operations, that nabla is in
all cases similar to a conventional vector. We shall use the expression
"multiplication by nabla" if, when nabla operates on any expression,
that expression is differentiated. Sometimes, nabla is multiplied by
a vector without operating on it as a derivative, in that case it is
applied to another vector [see (11.30), (11.32)].
Gradient. If we operate with V on a scalar 9, we obtain a vector
which is called the gradient of the scalar 9:
V9=i+J + k. (11.23)
Its components are:
V,9 = J, V, 9 J, V.94J (H24)
From equations (11.24), it can be seen that the vector V9 is per
pendicular to the surface 9 = const. Indeed, if we take a vector dl
lying on this surface, then, in a displacement dl, 9 does not change.
This is written as
> (11.25)
i.e., V9 is perpendicular to any vector which lies in the plane tangential
to the surface 9 = const, at the given point, which accords with oiir
assertion.
Differentiation of products. We now give the rules governing
differential operations with V.
First of all, the gradient of the product of two scalars is calculated
as the derivative of a product:
(11.26)
The divergence of a product of a scalar with a vector is calculated
thus :
div 9 A = (Vep, 9 A) + (V A , 9 A) = (AV ? ) + 9 (VA) 
=^A grad 9 + 9 div A. (11.27)
Here the indices 9 and A attached to V show what V is applied to.
Sec. 11] VECTOR ANALYSIS 99
We find the rotation of 9 A in a similar manner:
rot 9 A= [V 9 , 9 A] + [V^, 9 A] = [grad 9, A] + 9 rot A . (11.28)
Now we shall operate with V on the product of two vectors:
div [AB](V [AB]) = (V^ [AB]) + (Vu [AB]).
We perform a cyclic permutation in both terms, since V can be
treated in the same way as an ordinary vector. In addition, we put B
after VB in the second term, and here, as usual, we must change
the sign of the vector product. The result is
div [AB] = (B [V A A]) (A [V B B])=B rot A A rot B. (11.29)
Let us find the rotation of a vector product. Here we must use the
relationship [A[BC]] B (AC) C (AB):
rot [AB] = [V A [AB]] + [V B [AB]] = (V A B) A (V A A) B + ( V B B) A
(AV B ) B = (BV) A B div A + A div B (AV) B. (11.30)
Here we note the new symbols (BV) and (AV) operating on the
vectors A and B. Obviously, (AV) and (BV) are symbolic scalars,
equal, by definition of V, to
(AV) = A, V x + A y V Y + A,V, = A x ^ + A r ^ + A, J , (11.31)
and similarly for (BV). Then, (AV) B is a vector which is obtained
by application of the operation (11.31) to all the components of B.
Of the operations of this kind, we have yet to calculate grad AB :
grad (AB) =V A (AB) +V B (AB).
We use the same transformation as in the preceding case:
grad (AB) = (BV A ) A + [B [V A A]] + (AV B ) B + [A [V B B]] =
= (BV) A + (AV) B + [B rot A] + [A rot B]. ( 1 1.32)
Certain special formulae. We note certain essential cases of
operations involving V.
From the definition of divergence (11.7), we obtain from (11.27)
and (11.9)
div^= 1 divr + r grad = ~  ^  . (11.33)
Further,
and in general
rotr = 0. (11.34)
100 KLECTBO DYNAMICS [Part II
We now take
and for all components of r at once
(AV)r = A. (11.35)
In addition, we apply V to a vector depending only on the absolute
value of the radius vector. We note first of all that
fCf. (3.3.), where 1/r is differentiated], so that
Vr = y. (11.36)
Using the rule for differentiating a function of a function, we have
div A (r) = . Vr = AL. (11 . 37)
Here A is a total derivative of A (r) with respect to the argument /*,
i.e., a vector whose components are the derivatives of the three
components of A (r) with respect to r: A x , A v , A s .
Further,
rotA<r)=ArJJ] = J^. (11.38)
Repeated differentiation. Let us investigate certain results con
cerning repeated operations with V.
The rotation of the gradient of any scalar is equal to zero:
rot grad 9   "V,V?I = [VVJ 9=0, (11.39)
since the vector product of any vector (including V) by itself is equal
to zero. This can also be seen by expanding rot grad 9 in terms of its
components. The divergence of a rotation is also equal to zero:
div rot A = (V [VA]) = ([VV] A)=0 . (11.40)
Let us write down the divergence of the gradient of a scalar 9 in
component form. From equations (11.7) and (11.24) we have
divgrad 9 HVV)<p=8+ S + SACP. (11.41)
Here A (delta) is the socalled Laplacian operator, or Laplaciaii:
A = *L+ a2 _ + _^_
dx* ~ r dy* "" dz* '
Sec. 11]
VECTOR ANALYSIS
101
Filially, the rotation of a rotation can be expanded as a double
vector product:
rot rot A [V [VA]]=V (VA) (VV) Agrad div A A A . (11.42)
The last eqiiation can be regarded as a definition of AA. In curvi
linear coordinates, A'cp and A A are expressed differently.
Curvilinear coordinates. We shall further show how the gradient,
divergence, and rotation, as well as A of a scalar appear in curvilinear
coordinates.
Curvilinear coordinates q l9 q 2 , q 3 are termed orthogonal if only the
quadratic terms dql, dql, dql appear in the expression for the
element of length dl 2 , and not the products dq^ dq 2 , dq l dq& dq% rf</ 3 ,
similar to the way that dl 2 = dx 2 + dy 2 + dz 2 in rectangular coordinates.
In orthogonal coordinates
dl 2 = fil dql + hl dq\\hl dql . (11.43)
For example, in spherical coordinates q r = r, q 2 = $,q 3 = cp. The element
of length is
so that
sn
Let us construct an elementary parallelepiped (Fig. 20). Then the
components of the gradient will be
C f
d q 3
C
(11.44)
In order to find the divergence we repeat
the proof of the GaussOstrogradsky theorem
for Fig. 20. The area ADCB is equal to h 2 \
vector A through it is
Fig. 20
i dq 2 dq%. The flux of
Here, A 2 and A 3 are also taken for a definite value of q v The sum of the
fluxes through the areas ADCB and A'B'C'D' is
where we have used the expansion of the quantity h 2 h% A at the
point q l + dq t in terms of dq l9 jn a way similar to (11.4). The total flux
across all the boundaries is
102 ELECTRODYNAMICS [Part II
dj ^ ?7 (h * h * AI) + r (AS hi AZ) + ~i (AI * ^ s) dqi dq * dq * *
Let us now take advantage of the definition of divergence (11.8):
dJ div A /4 A 2 ^3 ^?i ^2 d# 3 =div Adi V .
Hence,
(11.45)
If, instead of A v A%, A 3 , we substitute the expressions (11.44), the
result will be the Laplacian of a scalar in orthogonal curvilinear coor
dinates. Thus, in spherical coordinates it is
With the aid of Stokes' theorem, we can also calculate the rotation
in curvilinear coordinates. We shall give it for reference without
proof:
, . l / d A 7 8 f
rotl A =  A * 3 ~ '* 2
ro D
,
rot
3
A ~~~" i " 7"*" l ~~/\ ** i ^i o ^* 1 ' t'o l
Mi \ 0ft ^^i 3 3 /
. I I d A i B 4 j \
A = ,, \ ~A~h  .4 T //! .
M \ a ?i a a J V
(11.47)
Exercises
Whore (from the requirements of the problem) expressing in terms of coordi
nates is not demanded, it is recommended that only the vector equations of the
present section (11.2ti)(11.42) be used.
1) Calculate the expressions: Answers:
a) A (r^O). A =divgrad *=* div = 0.
b)div9(r)r, rot 9 (r) r . 89 4 79; 0.
c) V (Ar), TAC A = const. At
d)V(A(r)r). A +l(rA).
e) div 9 (r) A (r), rot 9 (r) A (r) . 1 (rA) + 1 (pA) . _t [rA j + f [rA] t
f ) div [r [Ar]], A = const. 2 (Ar) .
g) rot [r [Ar]], A = const. 3 [ rA ] .
h)AA(r) [CM. (11,42)]. i + ? JL
r
i) V(A(r)B(r)). 1
r
j) rot[Ar], A = const. 2 A.
k) div [Ar], A  const.
DA!. ^a
r r>
See. 11] VECTOR ANALYSIS 103
2) Write down A^ in cylindrical coordinates.
3) Write down the three components of AA in spherical coordinates.
4) Two closed contours are given. The radius vector of points of the first
contour is r x , of the second contour, r>. The elements of length along each
contour are dl^ and d\ 2 , respectively. Prove that the integral
is equal to zero, 4 TT, 8 w, and n4 it, depending upon how many times the first
contour is wound round the second, linking up with the latter. v\ denotes diffe
rentiation with respect to r x (Ampere's theorem).
Changing the order of integration and performing a cyclic permutation of the
factors, we have
We apply Stokes' theorem (11.19) to the integral in d} l :
Wo use equation (11.30) ; rotj denotes differentiation with respect to tho compo
nents r x ; and d\ 2 in such a differentiation may be regarded as a constant vec
tor:
In
to
accoi
zero.
rot x 1 '
L Vl l',
'dance
Tliere
with
rerru
 . 2J '
exercise
ains, ther
, di.]
\ifi 2 v
la, the
efore,
(dl a Vi
if vi 1 r r 1 "
last term containing A t 
\ * /ji *,
. v
1 l r r
1
equal
1':
i)
iM 1S
i
T7
Vl 'i
 r 2l
)Vl rxr a 
Vi I r r
'2! '
since a function of the difference T I r. 2 is differentiated.
For short, we write r=r x r 2 . Then the required integral will be
* = J (d\ 2 V 2 ) J <*Si Vi  ri _ ra  = J (dl* Vi) J ^i V
We shall now explain tho geometrical sense of the second integrand, i.e.,
d^ v =   L ^ y . Tho scalar product    is the projection of
an element of the surface d^ pulled over the first contour, on the radius vector
r drawn from a point on the second contour. In other words,    is equal
to the projection of the area da on a plane perpendicular to r. This projection,
divided by r 2 , is equal to the solid angle d& at a point r 2 on the second contour
subtended by the area ds r The integral I  3  is, therefore, that solid angle
n which is obtained if a cone is drawn with vertex at the point r 2 , so that the
generating line of the cone formed the contour l t .
The differential (d\ 2 V 2 ) Q is the increment of solid angle O obtained in
shifting along the contour 1 2 a distance d\ 2 . Thus,
J (d\ 2 V 2 ) Q
The integral of this quantity around a closed contour is equal to the total
change in solid angle in traversing the contour 1 2 . Let the initial point of cir
cumvention lie on the surface x . Then the solid angle subtended by the surface
104 ELECTRODYNAMICS [Part TT
at the origin is 2 TC. If the contours aro linked, then the solid angle will be
2 T: after the circumvention, since the ure,a is observed from a terminal point
on the other side. If the contours are riot linked, then the solid angle is once
again its initial value, 2 TC, arid the integral is equal to zero. Thus, when the
contours are linked n times, the integral in d\ 2 is equal to 4 TC n.
Sec. 12. The Electromagnetic Field. Maxwell's Equations
Interaction in mechanics and in electrodynamics. The interaction
of charged bodies in electrodynamics is principally an interaction of
charges with an electromagnetic iield. However, the physical concept
of the iield in electrodynamics differs essentially from the field concept
in Newtonian mechanics.
We know that the space in which gravitational forces act is called a
gravitational iield. The values of these forces at any point of the Held
is determined, in Newtonian mechanics, by the instantaneous positions
of the gravitating bodies, no matter how far they are from the given
point. In electrodynamics, such a iield representation is not satis
factory : during the time that it takes an electromagnetic disturbance
to move from one charge to another, the latter can move a very great
distance. Elementary charges (electrons, protons, mesons) very often
have velocities close to the velocity of propagation of electromagnetic
disturbances.
Modern gravitational theory (the general theory of relativity, sec
Sec. 20) shows that gravitational interaction, too, propagates with a
finite velocity. But since macroscopic bodies move considerably
slower, within the scale of the solar system, the finite velocity of
propagation, of gravitational forces introduces only an insignificant
correction to the laws of motion of Newtonian mechanics.
Tn the electrodynamics of elementary charges, the finite velocity of
propagation of electromagnetic disturbances is of fundamental signi
ficance. When speaking of point charges, the action of a field on the
charge is always determined only by the field at the point where the
charge is located, and only at the instant when the charge is at this
point. As opposed to the "action at a distance" of Newtonian mecha
nics, such interactions are termed "shortrange."
If the energy or momentum of a charged particle is changed under
the action of a field, they can be imparted directly only to the electro
magnetic field, since a finite interval of time is necessary for the energy
and momentum of other particles to be changed. But this means that
the electromagnetic field itself possesses energy and momentum,
whereas in Newtonian mechanics it was sufficient to assume that only
the interacting particles possessed energy and momentum. It follows
from this that the electromagnetic field is itself a real physical entity
to exactly the same extent as the charged particles. The equations of
electrodynamics must describe directly the propagation of electro
See. 12] THE ELECTROMAGNETIC FIELD. MAXWELL \S EQUATIONS 105
magnetic disturbances in space and the interaction of charges with the
field.
Interaction between charges is effected through the electromagnetic
field. Such laws as the Coulomb or BiotSavart laws (in which only
the instantaneous positions and the instantaneous velocities of the
charges appear) are of an approximate nature and are valid only when
the relative velocities of the charges are small compared with the
propagation velocity of electromagnetic disturbances.
It will be shown later that this velocity is a fundamental constant
which appears in the equations of electrodynamics. It is equal to the
velocity of light in vacuo and, to a high degree of precision, is
3 x IQ IQ cm/sec.
A field in the absence of charges. The independent reality of the
electromagnetic field is particularly evident from the fact that electro 
dynamic equations admit of a solution in the absence of charges.
These solutions describe electromagnetic waves, in particular light
waves, in free space. Thus, electrodynamics has shown that light is
electromagnetic in nature.
In the course of two centuries, the protagonists of the wave theory
of light considered that light waves were propagated by a special
elastic medium permeating all space, the socalled "ether." In order
to represent the spread of oscillations it was, naturally, necessary to
have something oscillating. This "something" was called the ether.
Proceeding from an analogy with the propagation of sound waves
in a continuous medium, the ether was endowed with the properties
of a fluid, physical phenomena being explained simply by reducing
them to definite mechanical displacements of bodies. In particular,
light phenomena were regarded as displacements of particles of the
special medium, the ether.
In this, a peculiar "abhorrence of a vacuum" was apparent or, more
exactly, a purely speculative representation of empty space where
"nothing exists" and, hence, where nothing can occur. Physicists did
not at once come to realize that the electromagnetic field itself was
just as real as the more tangible "ponderable matter." Electrodynamic
laws are those elementary concepts, from which the interaction of
atoms should be deduced, which interaction accounts for the proper
ties of real fluids that are incomparably more complicated than the
properties of a field in "empty space/' i.e., in the absence of charges.
There is no sense in reducing a field to an imaginary fluid merely in
order to avoid the idea of "empty space." Physical space is the carrier
of the electromagnetic field and is, therefore, inseparable from the
state and motion of real objects. As regards the term "ether," which
still persists in the field of radio, it expresses nothing other than the
electromagnetic field.
The electromagnetic field. Let us now establish the basic equations
of electrodynamics. We shall proceed from certain elementary laws,
106 ELECTRODYNAMICS [Part II
which we assume the reader knows from a general course of physics
or electricity. These laws will first be used in the absence of matter
consisting of atoms or, as is usually said in electrod3^namics, in the
absence of a "material medium." By this term we must not under
stand any encroachment on the material nature of the electromagnetic
iield itself. From the electrodynamic equations for free space we shall,
later on, derive the equations for an electromagnetic field in a medium
(a conductor or dielectric).
As is known, the electromagnetic field in a medium is described by
four vector quantities: the electric field, the electric induction, the
magnetic iield, and the magnetic induction. The force acting on unit
electric charge at a given point in space is called the electric field
intensity. In future, instead of the field intensity, we shall simply
speak of the field at a given point in space. The magnetic field intensity
or, for short, the magnetic field is defined analogously. Separate magne
tic charges, unlike electric charges, do not exist in nature; however,
if we make a long permanent magnet in the form of a needle, then the
magnetic force acting at its ends will be the same as if there existed
point charges at the ends.
A rigorous definition of the electric and magnetic induction vectors
will be given in Sec. 1C, where the field equations in a medium will be
derived from the equations for point charges in free space. It need
only be recalled that in free space there is no need to use four vectors
for a description of the electromagnetic field, only two vectors being
sufficient: the electric and magnetic fields.
System of units. We shall consider that all electromagnetic quanti
ties are expressed in the CGSE system, i.e., in the absolute electro
static system of units. In this system the dimensions of electric charge
are gnW cm 3 //sec and the dimensions of field intensity, both electric
and magnetic, are gmVi/cmVa see. If we substitute charge, expressed
in this .system of units, into the equation for Coulomb's law, then the
interaction force between charges is expressed in dynes (gm cm/sec 2 ).
Electromotive force. Let us recall the definition for electromotive
force in a circuit : this is the work performed by the forces of the
electric field when unit charge is taken along the given closed circuit.
And it is absolutely immaterial what the given circuit represents:
whether it is filled with a conductor or whether it is merely a closed
line drawn in space. Let us write down the expression for electromotive
force (abbreviated as e. m. f.) in the notation of Sec. 11. The force
acting ou unit charge at a given point is the electric field E. The work
done by this force on an element of path d\ is the scalar product Ed 1.
Then, the work done on the whole closed circuit, or the e. m. f., is
equal to the integral
e.m.f. j*Edl. (12.1)
Sec. 12] THE ELECTROMAGNETIC FIELD. MAXWELL'S EQUATIONS 107
Magneticfield flux across a surface. Let us suppose that some
surface is bounded by the given circuit. We shall denote the magnetic
field by the letter H. The magneticfield flux through an element of
the chosen surface is, by the definition given in Sec. 11, d O HdS.
The magneticfield flux through the whole surface, bounded by the
circuit, is
O = fHds. (12.2)
It can be conveniently represented thus. Let us consider a section
of the surface through which unit flux A * = 1 (in the CGSE system)
passes. We draw through this section of the surface a line tangential
to the direction of the field at some point on the surface. A line which
is tangential to the direction of the field at its points is called a magnetic
line of force. For this reason, the total flux 4 is equal, by definition,
to the number of magnetic lines of force crossing the surface.
Magnetic lines of force are either closed or extended to infinity.
Indeed, a magnetic line offeree may begin or end only at a single charge,
but separate magnetic charges do not exist in nature. In a permanent
magnet the lines of force are completed inside the magnet.
From this it follows that a magnetic flux through any surface,
bounded by a circuit, is the same at a given instant. Otherwise, a
number of the magnetic lines of force would have to begin or end in
the space between the surfaces through which different fluxes pass.
Consequently, at a given instant, a constant number of magnetic
lines of force, i.e., a constant magnetic field flux passes across any
siirface bounded by the circuit. Therefore, the flux can be ascribed
to the circuit itself, irrespective of the surface for which it is calculated.
Faraday's induction law. Faraday's induction law is written in the
form of the following equation:
e.m.f. i. (12.3)
If all the quantities are expressed in the CGSE system, then the
constant of proportionality c is a universal constant with the di
mensions of velocity equal to 3 x 10 10 cm/sec.
Usually, Faraday's law is applied to circuits of conductors ; however,
e. in. f. is simply the quantity of work performed by unit charge in
moving along the circuit, and, for a given field value through the
circuit, cannot depend upon the form of the circuit. The e. m. f. is
simply equal to the integral f Edl. In a conducting circuit, this work
can be dissipated in the generation of Joule heat ("an ohmic load").
However, it is completely justifiable to consider the circuit in a vacuum
also. In this case, the work performed on the charge is spent in increas
ing the kinetic energy of the charged particle, as, for instance in the
case in an induction accelerator, the betatron.
1(>8 0LECTRODYNAMICS [Part II
Maxwell's equation 'fop rot E. Thus, equation (12.3) refers to any
arbitrary closed circuit. WeliuTwtitute the definitions (12.1) and (12.2)
into this equation:
(12.4)
The lefthand side of the equation can be transformed by the
Stokes theorem (11.19) and, on the righthand side, the order of the
time differentiation and surface integration can be interchanged, since
they are performed for independent variables. In addition, taking
this integral over to the lefthand side, we obtain
8 = 0. (12.5)
But, the initial circuit is completely arbitrary, i.e., it can have
arbitrary magnitude and shape. Let us assume that the integrand, in
parentheses, of (12.5) is not equal to zero. Then we can choose the sur
face and the circuit that bounds it so that the integral (12.5) does not
become zero. Thus, in all cases, the following equation must be satis
fied:
rotE + y a jJ=0. "' (12.6)
Tn comparison with (12.3), this equation does not contain anything
new physically ; it is the same induction law, but rewritten in differen
tial form for an infinitely small circuit (contour). In many applications
the differential form is more convenient than the integral form.
We shall see later that the constant c is equal to the velocity of
light in free space. ?
The equation for 4'iv II. As we have already said, magnetic lines of
force are either closed or go off to infinity. Hence, in any closed surface,
the same number of magneticfield lines enter as leave. The magnetic
field flux in free space, across any closed surface, is equal to zero:
f*Hds0. (12.7)
Transforming this integral to a volume integral according to the
GaussOstrogradsky theorem (11.6), we obtain
>. (12.8)
Due to the fact that the surface bounding the volume is completely
arbitrary, we can always choose this volume to be so small that the
integral is taken over the region in which div H has constant sign if
it is not equal to zero. But then, in spite of (12.7) and (12.8), j div H d F
will not be equal to zero. For this reason, the divergence of H must
become zero:
vSeC, 12J THE ELECTROMAGNETIC FIELD. MAXWELL'S EQUATIONS 109
divH = 0. (12.9)
(12.9) is the differential form of (12.7) for an infinitely small volume.
In Sec. 11 it was shown that the divergence of a vector is the density
of sources of a vector field. The sources of the field are free charges
from which the vector (force) magneticfield lines originate. Thus,
(12.9) indicates the absence of free magnetic charges.
Equations (12.6) and (12.9) are together called the first pair of
Maxwell's equations. ,M
Let us now introduce the v second pair.
The equation for (UvJE.. The electricfield flux through a closed surface
is not equal to zero, but to the total electric charge e inside the surface
multiplied by 4 TC (Gauss' theorem) :
v Jli . fEds = 47ue. v (12.10)
This theorem is derived from Coulomb's law for point charges.
The field due to a point charge v is expressed by the following equation:
Here, r is a radius vector drawn from the point situated at the charge
to the point where the field is defined. The field is inversely propor
tional to r 2 and is directed along the radius vector.
Let us surround the charge by a spherical surface centred on the
charge. The element of surface for the sphere rfs is r 2 dl , where
d Q is an elementary solid angle and indicates the direction of the
normal to the surface. The flux of the field across the surface element is
The flux across the whole surface of the sphere is I edl^=e dfc^
~4 TT e. But since lines of force begin only at a charge, the flux will
be the same through the sphere as through any closed surface around
the charge. Therefore, if there is an arbitrary charge distribution e
inside a closed surface, then equation (12.10) holds.
In order to rewrite this equation in differential form, we introduce
the concept of charge density. The charge density gjs the charge con
tairied in unit volume, so that the total charge in the volume is related
to the density by the following equation:
e^lpdF. (12.11)
Ae
Hence, p=lim ^ . Introducing the charge density in (12.10), we
obtain AF>O
110 ELECTKODYNAMICS [Part II
0. (12.12)
Repeating the same argument for this integral as we used for (12.8),
we have
divE=4jcp. (12.13)
According to (11. 8) we can say that the density of sources of an
electric field is equal to the electric charge density multiplied by 4 ru.
The density function for point charges. The density function for
point charges is obtained by a limiting process. Let us initially assume
that a iinite quantity of charge is distributed in a small, but finite,
volume A V. Then p must be regarded as the ratio ^ .
If we let the volume A V tend to zero, then the density will have a
very peculiar form: it will turn out to be equal to zero everywhere
except at the place where the charge is situated, and at that point it
will convert to infinity, since the numerator of the fraction rrr is
finite and the denominator is infinitely small. However, the integral
remains equal to the charge e itself. Thus, the concept of charge density
can also bo used in the case of a point charge. In. this case, p is under
stood to be a function which is equal to zero everywhere except at the
point of the charge. The volume integral of this function is either equal
to the charge e itself, if the charge is situated inside the integration
region, or zero, if the charge is outside the region of integration.
The charge conservation law. One of the most important laws of
electrodynamics is the law of conservation of charge. The total charge
of any system remains constant if no external charges are brought into
it. In. all charge transformations occurring in nature, the law of con
servation of charge is satisfied with extreme precision (while the law
of conservation of mass is approximate!).
In order to formulate the chargeconservation la\v in differential
form, we must introduce the concept of current density. This vector
quantity is defined as
j = pv, (12.14)
where v is the charge velocity at the point where the density p is
defined. The dimensions of charge density are charge/cm 3 , and of
current density, charge/cm 2 sec (i.e., the dimensions of charge pass
ing in unit time through unit area). For a point charge, v denotes
its velocity and p the density function defined above.
The total current emerging from an area is
/ = [jds= fp(vds). (12.15)
Sec. 12] THE ELECTROMAGNETIC FIKLD. MAXWELL'S EQUATIONS 111
It must be equal to the reduction of charge inside the surface in
unit time, i.e.,
l = ~ (12.16)
(the chargeconservation law in integral form). Substituting e from
(12.11) and transforming / by the GaussOstrogradsky theorem,
we obtain
F=0. (12.17)
Since the volume, over which the integration is performed is arbitrary,
the conservation law for charge in differential form follows from
(12.17):
^ + divj =  + divpv = 0. (12.18)
Displacement current. From directcurrent theory, it is known
that current lines are always closed. Indeed, open circuited lines
indicate that there is either an accumulation or deficiency of charge
at their ends. But we can also define vector lines such that they
will always be closed (or will have to go off to infinity) in the case
of alternating currents. For this we substitute the derivative ? ,
according to (12.13), into the equation of the chargeconservation
law (12.18). This derivative is equal to div^ . Hence, we
always have the relation
Comparing (12.19) and (12.9), we see that the vector lines
. 1 3E
J " 47c dt
are always closed. The vector ~ is called the displacement cur
rent density. Together with the charge transport
current j, the displacement current forms a closed
system of current lines.
The displacement current can be more vividly
demonstrated in the following way. Fig. 21 shows
a capacitor whose plates are joined by a conductor.
The current / flowing in the conductor is equal
to the change of charge on the plates in unit ~.
time : j __ &e_ lg *
But the charge on the plates is related to the field in the capacitor
by the relationship
112 ELECTRODYNAMICS [Part II
where / is the area of the plates. Whence
r _ f 2K
47T ~dt '
Consequently, the quantity   7 can be interpreted as the density
of some current which completes the conduction current, of density
7 =  This corresponds to the more general equation (12.19).
The conception that the magnetic action of a displacement current
does not difter from the magnetic action of ordinary current is basic
to MaxweJlian electrodynamics.
Magnetomotive force. By analogy with electromotive force f Edl,
r
we deli ne the magnetomotive force 11 e/1, where the integration
is performed over a closed circuit. Using the BiotSavart law for
direct currents, it may be shown that the magnetomotive force in
a closed circuit is equal to the electric current, /, crossing a surface
47C
bounded by the circuit, multiplied by  In other words,
(12.20)
This relationship can be shown most simply by assuming that
a direct current / is flowing through an infinitely long straightline
Circuit. We shall calculate the magnetomotive force in a circuit of
circular form, the current line passing through the centre of the
circle perpendicular to its plane. The magnetic field is tangential
to the circle and equal to // " r in accordance with the BiotSavart
law, where r is the radius of the circle. Thus, for the circuit we have
chosen, the absolute value of II is constant and the magnetomotive
force is equal to 2nrH^ ^ .
For a circuit of arbitrary form, we should use the BiotSavart
law in differential form:
Then the magnetomotive force is represented by the integral
However, in accordance with Ampere's theorem (exercise 4, Sec. 11),
this integral is equal to if the circuit in which the magneto
motive force is calculated is linked with the currentcarrying circuit.
Sec. 12] THE ELECTROMAGNETIC FIELD. MAXWELL'S EQUATIONS 113
The equation H&r. EQlLH. Following Maxwell, we shall assume that
equation (12.20) is also true for displacement current if the lield
is variable. The current lines will then always be closed, and in cal
culating magnetomotive force we can use all the arguments that
we have used for continuous current. In the case of varying fields
and currents, /, in the formula for magnetomotive force, denotes
the total current passing through the circuit, i.e., the sum  j ds + ~r
jrfds. Naturally, this assumption is not obvious beforehand
and is justified by the fact that Maxwell's equations provide for the
explanation or prediction of the entire assemblage of phenomena
relating to rapidly changing electromagnetic fields (the displacement
current does not usually exist for slowly varying fields).
Let us assume that the total current is formed by the current
produced by charge transport (of density equal to j) combined with
a displacement current with density .  (  . In accordance with
our assumption,
(12.21)
Transforming the lefthanJ side of (12.20) in accordance with
Stockcs' theorem (11.19) and combining with /, we obtain
Applying the same argument to eqiiation. (12.22) as applied to
(12.5), we arrive at the differential equation
rotll 1J?A_1^L=0. (12.23)
c dt c v '
It is easy to see that this equation agrees with the law of charge
conservation. Indeed, we operate on it by div. According to (11.40),
div rot H =0, so that we are left with
Substituting div E from (12.13), we arrive once again at (12.18), i.e.,
the law of conservation of charge.
Equation (12.23) is not merely an expression for the BiotSavart
law in differential form. In (12.23), we have introduced the displace
ment current, which is not involved in the theory of continuous
currents.
The Maxwell system of equations. Let us once again write down
the system of Maxwell's equations for free space.
80060
114 ELECTRODYNAMICS [Part II
The first pair:
divH = 0. (12.25)
The second pair:
(12.27)
In these equations we consider p and j, i.e., the charge and current
distributions in space, to he known. The unknowns, to be determined,
are the fields E and II. Each of them has three components.
In spite of the fact that both pairs form, together, eight equations,
only six of them are independent, according to the number of field
components. Indeed the three components of each rotation are
constrained by div rot = and, hence, are not independent of one
another.
Electromagnetic potentials. We can introduce new unknown quan
tities such that each equation, will contain only one unknown. In
tins way the overall number of equations is reduced. These new
quantities are called electromagnetic potentials.
We choose the potentials so that the first pair of Maxwell's equations
are identically satisfied, hi order to satisfy equation (12.25), it is
sufficient to put
II rot A, (12.28)
where A is a vector called the vector potential. Then, according to
(11.40), the divergence of II will be equal to zero identically. We
shall look for the electric field in the form:
where 9 is a quantity called the scalar potential.
From (11.39) rot^Vcp=0. Substituting (12.28) and (12.29) in
(12.24), we obtain the identity.
The detcrminacy of potentials. The electromagnetic fields E and H
are physically determinate quantities since, through them, the forces
acting on charges and currents can be expressed. The fields are ex
pressed in terms of potential derivatives. Therefore, potentials are
determined only to the accuracy of the expressions that cancel in
differentiation. These expressions shoiild be chosen so that the po
tentials satisfy equations of the simplest form. We shall now find
the most general potential transformation which does not change the
fields.
From equation (12.28) it can be seen that if we add the gradient
of any arbitrary function to the vector potential, the magnetic
Sec. 12] THE ELECTROMAGNETIC FIELD. MAXWELL'S EQUATIONS 115
field will not change, since the rotation of a gradient is identically
equal to zero. Putting
A = A'+V/(*,y,M), (12.30)
we see that the magnetic field, expressed in terms of such a modified
potential, remains unchanged:
In order that the addition of V/ should not affect the electric field,
we must also change the scalar potential:
? = <P'4ff, (12.31)
where / is the same function as in (12.30). Then, for the electric
field, we obtain
_._
_ i a A' _ ,
~~~TTr~~ V(p
Consequently, the electric field does not change either. Thus,
the potentials are determined to the accuracy of the transformations
(12.30), (12.31), which are called gauge transformations.
The Lorentz condition. Let us now choose an arbitrary function /
such that the second pair of Maxwell's equations leads to equations
for the potentials of the simplest possible form. Substituting (12.28)
and (12.29) in (12.26) gives
rotrotA=^jj>V 9 + fi, (12.32)
We express rot rot A with the aid of (11.42). Then (12.32) is reduced
to the following form:
We shall now try to eliminate the quantity inside the brackets.
We denote it, for brevity, by the letter a, and we perform the trans
formations (12.30) and (12.31) on the potentials. Then the quantity
a is reduced to the form
=divA + 14f=divA' + 4f + A/ig. (12.34)
The function / has, so far, remained arbitrary. Let us now assume
that it has been chosen so as to satisfy the equation
A/ = . (12.35)
116 ELECTRODYNAMICS [Part II
Then, from (12.34), it is obvious that the potentials will be subject
to the condition
~4r7^ ( 12  36 )
This is called the Lorentz condition.
As was shown, the expression of fields in terms of potentials is
not changed by a gauge transformation. For this reason we shall
always consider, in future, that this transformation is performed
HO that the Lorentz condition is satisfied; the primes in the potentials
can then bo omitted.
The equations for potentials. From the Lorentz condition and
(12.33), we obtain the equation for a vector potential:
It is now also easy to obtain the equation for a scalar potential.
From (12.27) we have
i P
div E =  TTT div A A9 = **?
C f)t
Substituting div A from the Lorentz condition (12.36), we obtain
(12.38)
Equations (12.37) and (12.38) each contain only one unknown.
Therefore, each equation for potential does not depend on the rest
and can be solved separately.
The equations for potential are second order with respect to coor
dinate and time derivatives. For a solution, it is necessary to give
not only the initial values of the potentials, but also the initial values
of their time derivatives.
Gauge invarianco. As we shall see later, especially in the following
section and in Sec. 21, which is devoted to the motion of charges
in an electromagnetic field, it is necessary, in many cases, to use
equations involving potentials. But, since potentials are ambiguous,
we must take care that the form of any equation involving potentials
does not change under gauge transformations (12.30) and (12.31),*
since such transformations involve a completely arbitrary function
/ which can be chosen to be of any form. It is clear that no physical
result can depend on the choice of this function, i.e., on an arbitrary
gauge transformation. In other words, equations involving potentials
must bo gauge invariant.
* This does not refer to equations (12.37) and (12.38), from which the poten
tials are determined in accordance with the condition (12.36).
Sec. 13] THE ACTION PRINCIPLE FOB THE ELECTROMAGNETIC FIELD 117
Sec. 13. The Action Principle for the Electromagnetic Field
The variational principle lor the electromagnetic field. In the first
part of this book it was shown that the equations of mechanics,
obtained from Newton's laws (Sec. 2), lead to the principle of least
action (Sec. 10). We obtained the equations of electrodynamics in*
the preceding sectioi} by proceeding from certain simple physical
laws and the assumption about the magnetic effect due to displace
ment current. In this section, Maxwell's equations will be reduced
to the variational principle, which is the principle of least action
for the electromagnetic field, i
Electrodynamics is not equivalent to the mechanics of particle
systems or to the mechanics of liquids, which are based on Newton's
laws. All the same, to a very considerable extent, electrodynamical
laws are analogous to the laws of mechanics. This analogy can best
be seen from the principle of least action for the electromagnetic
field.
The variational formulation best of all allows us to derive the
conservation laws for the electromagnetic field. The corresponding
integrals of motion for a field coincide with the wellknown mechani
cal integrals energy, linear momentum, and angular momentum.
In a closed system consisting of charged particles and a field, the
total energy, total linear momentum, and total angular momentum
of the charges and field are conserved.
In this sense, electrodynamics is indeed "a dynamics" of the elec
tromagnetic field, though this by no means signifies that the laws
of electrodynamics can be obtained from Newton's laws. Both are
equivalent to certain integral variational principles, but the action
functions are, of course, of entirely different form.
It is a noteworthy fact that Maxwell at first tried to construct
mechanical models of the ether, but in his later work he rejected
them and obtained the general equations of electrodynamics by
means of a generalization of known elementary laws of electro
magnetism.
The Lagrangian function for a field. In order to formulate the prin
ciple of least action it is necessary to have an expression for the
Lagrangian. The choice of Lagrangian in mechanics is determined
by considerations based on the relativity principle of Newtonian
mechanics, which is formulated with the aid of Galilean transfor
mations (Sec. 8). As wiU be explained in detail in Sees. 20 and 21,
Galilean transformations are not valid in electrodynamics and are
replaced by the more general Lorentz transformations, based on
the Einstein relativity principle. These transformations allow the
Lagrangian for the electromagnetic field to be uniquely found;
this will be done in Sec. 21. In this section, the choice of Lagrangian
is justified by the fact that the already familiar Maxwell equations
118 ELECTRODYNAMICS [Part II
are obtained from it. Similarly, in Part I, the principle of least action
was formulated after Lagrarige's equations had been obtained on
the basis of Newton's laws. TJiis confirmed the truth of the integral
principle.
In finding the Lagrangian for a system of free particles, a summa
tion is performed over the coordinates of the particles. The electro
magnetic field, if we use the terminology of mechanics, is a system
with an infinite number of degrees of freedom because, for a complete
description of tho Held, we must know all its components at all points
of space, where they differ from zero. But the points of space form
a nondenumorabh? set, i.e., they cannot be numbered in any order,
For this reason, for tho electromagnetic field the summation in the
Lagrangian is replaced by an integration with respect to continuously
varying parameters, i.e., coordinates of points in which the field
is given. The point coordinates are analogous to the indices which
label the degrees of freedom of a mechanical system.
The equations of mechanics are second order in time with respect
to generalized coordinates (jk. The equations for potentials (12.37)
and (12.38) are also second order in time. Therefore, potential quanti
ties should be chosen as the generalized coordinates.
In other words, A (r, t), 9 (r, t) correspond to qi* (t), where A and 9
are potentials which are generalized coordinates of an electromagnetic
field. The value of the radius vector r for the point at which the po
tential is taken corresponds to the number of the generalized co
ordinate k.
In order to write down the complete Lagrangian function, we
must first of all defino it in an element of volume dV and integrate
over tho volume occupied by the field. It has already been mentioned
that in this section we will proceed immediately from a Lagrangian
that leads to correct Maxwell equations; the choice of this Lagrangian
as based on considerations related to the relativity principle will
be left to Sec. 21. The Lagrangian is of the following form:
Since tho potentials are liable to be generalized coordinates of the
field, expression (13.1) should be rewritten thus:
Here, in place of a summation over the degrees of freedom, an
integration over the volume has been performed.
Tho extremal property of action in electrodynamics. We shall now
show that action, i.e., S = I L dt, possesses the same variational
property in electrodynamics as it does in mechanics: its variation
Sec. 13] THE ACTION PRINCIPLE FOR THE ELECTROMAGNETIC FIELD 119
becomes zero if the field satisfies the correct equations of motion
(in this case, the Maxwell equations).
We shall begin with variation with respect to the scalar potential 9 :
V. (13.3)
As was shown in Sec. 10, variation and differentiation are com
mutative so that SV9 = V 8 9. According to (12.29) we replace
the term (~^ +V<p\ by E. Therefore,
We shall now make use of equation (11.27), in accordance with
which
E V $9 = div (89)  89 div E . (13.5)
We then obtain
=/ [ "^F *9 (^  P)] dV. (13.6)
The first term in (13.6) can be transformed into a surface integral,
so that $ L will have the form
F. (13.7)
We shall consider that the first integral is taken over a surface
on which & 9 becomes zero, similar to the way that, in Sec. 10, 8 q
was equal to zero at the limits of integration (a surface is the limit
for a volume integral).
Therefore,
(13.8)
However, since
(13.9)
[see (12.27)], S 9 L and hence 8 9 S, becomes zero as expected.
Let us now vary L with respect to A. This variation has the form
8AL= rrj_ (i^A  ls aA_ i rotASrotA JSA,
J L 4?r \ C d / C c>2 4rr c J
(13.10)
Once again we interchange the differentiation and variation signs
and, where possible, we replace the potentials by fields after variation.
We obtain S A L in the following form :
F. (13.11)
120 ELECTRODYNAMICS [Part 11
Let us write down the transformation by parts:
II rot 8 A :   div [USA] + 8 A rot II . (13.13)
The last equation follows from (11.29). To take advantage of
(13.12), we must write down the variation of the action $ instead
of the variation of L. Then the first term of (13.12) can be directly
integrated with respect to time, and 8^ S will be
t, 'i ii
8 A *S' (S^Ldt^ r  f ESAdK +/ [dt f[II8AJeZs +
A J A 4Kr, J ' 4rr J J L J
(13.14)
The variation S A is equal to zero at the initial and final instants
of time and t l9 as well as over the surface bounding the field.
Therefore,
A, (.3.15)
and since the field satisfied the equation
ro t.\l = ...+ *ZL (13.16)
[see (12.26)], &A ti is equal to zero.
The first pair of Maxwell's equations is, of course, satisfied identi
cally if the fields are expressed in terms of potentials in accordance
with (12.28) and (12.20).
Thus, Maxwell's equations can be interpreted as equations of the
mechanics of an electromagnetic field. They could be obtained from
the variational method, starting from the Lagrangian (13.1) and the
requirement that the variation of action should be equal to zero
for any arbitrary variations of the scalar and vector potentials.
For this it is sufficient to repeat the arguments set out in Sec. 10,
as applied to integrals (13.8) and (13.16).
The invariance of action with respect to a potential gauge trans
formation. We shall now show that action is invariant under gauge
transformations (12.30) and (12.31), despite the fact that it involves
not only fields, but also potentials contained in the last two terms
of equat ion (13.1). We shall call the corresponding part of the action S 1 :
^  P9 ). (13.17)
Sec. 13] THE ACTION PRINCIPLE FOR THE ELECTROMAGNETIC FIELD 121
Let us now apply gauge transformations (12.30) and (12.31) to
A and cp. This gives
dV  P9 < + + ? . (13.18)
We transform by parts terms containing /:
j V/= div (/j)  / div j , p { = jy (P/)  / /f
Substituting this in the integral /S^ and performing the integration,
as iu (13.14), we have
o
+ jdt j dV [4 3   P*'  /(div j + *J)  (13.10)
However, the integrated terms do not affect the Maxwell equations
since, when performing a variation of /S\, both S^ L and SA />
are equal to zero at the boundaries of the integration region. We
have already encountered this in (10.9). The term, proportional
to /, under the integral sign, is multiplied by the quantity div j +  p ,
which is identically equal to zero according to the charge conser
vation law (12.18)'. Thus. ^ retains the form (13.17).
The energy of a field. Maxwell's equations also apply to a free
electromagnetic field not containing charges or currents. For this
it is sufficient to omit from them the terms "* and 4 n p. in ac
c r
cordance with (13.1) and (13.2), the Lagrangian for a free field is
rdV. (1320)
We shah 1 now determine the energy of an electromagnetic field
by proceeding from the general equation (4.4). First, let it be recalled
that the values of potentials at all points of space arc generalized
coordinates. But then the derivatives  are generalized velocities.
o t
Consequently, the expression
reduces to the form
by means of the comparison
122 ELECTRODYNAMICS [Part II
We shall now show that energy is expressed only in terms of the
field, and not in terms of potentials. Using (12.29) we write down
the energy thus:
This expression is not invariant with respect to a gauge transformation
and must be transformed.
Transforming the term EV<? by parts, we have, from (11.27),
EV? = div (9) 9divE ~div9E,
since div E =0 for a field free of charges. The volume integral of
div rp E is transformed into a surface integral. However, according
to the meaning attached to L and $ , the integration should be
performed over the whole region occupied by the field (this is analog
ous to summation over all the degrees of freedom of the system).
At the boundary of this region, the field is equal to zero by definition,
so that the surface integral in the expression for energy also becomes
zero. From this we obtain the required expression for the energy
of an electromagnetic field in the absence of charges:
$ =  1  J (E 2 + II 2 ) A V. (13.21)
Hence, the quantity
=J^l (13.22,
may be interpreted as the energy density of the electromagnetic
field. It is invariant with respect to a gauge transformation of a
potential.
Conservation of the total energy of field and charges. We shall
now show that the energy $ (13.21), together with the energy of the
charges contained in the field, is conserved, i.e., g is the energy in the
usual, mechanical, sense of the word, and not some quantity which
is formally analogous to it only as regards its derivation from the
variatioual principle.
To do this wo multiply equation (12.26) scalarly by E and (12.24)
scarlarly by H, and subtract the second from the first. This gives
the following relationship:
1 /T^ aE IT ^ H \ V
E 5 + H ^r = E
c \ dt dt I
V 4U TT 4.TI
E rot H H rot E
Now taking advantage of equation (11.29), we reduce the equation
obtained to the form
p V
pvE. (13.23)
Sec. 13] THE ACTION PRINCIPLE FOR THE ELECTROMAGNETIC FIELD 123
Here we have put j = p v by definition. We now integrate (13.23)
over some volume, though not necessarily the whole volume oc
cupied by the field, and transform the integral of div to a surface
integral :
(13.24)
Let us first consider the second integral on the right. By definition,
the quantity p d V is the charge element de. The product E d e is
the electric force acting on this charge element. The scalar product
d e E v = d e E , is equal to the work done on the element
of charge in unit time or put in another way to the change in
kinetic energy T of the charge in unit time. Later, we shall show
that the magnetic field does not perform work on charges (Sec. 21).
To summarize, equation (13.24) can also be written as follows [the
dT
last integral in (13.24) will be represented in the form T , i.e.,
the work done in unit time] :
~ ( + T)  J^ [EH] dV . (13.25)
The Poynting vector. Thus, the decrease in energy, in unit time,
of an electromagnetic field and of the charged particles contained
therein is equal to the vector flux '  [E H] across the surface bound
ing the field. If this surface is infinitely distant and the field on it
is equal to zero, what we have is simply the energy conservation
law for an electromagnetic field and for the charges within it. Other
wise, if the volume is finite, the righthand side of equation (13.25)
indicates what energy passes in unit time through the surface bound
ing the volume. Hence, the quantity
U = ^[EH] (13.26)
represents the energy crossing unit area in unit time or, more simply,
the energy density flux vector (the Poynting vector).
Field momentum. Similar computations, which we shall not give,
show that an electromagnetic field possesses momentum. The mo
mentum of a field is given by the following integral:
1 [EH]dF. (13.27)
If the electromagnetic field interacts with some obstacle, for
example, the walls of the enclosure in which it is contained, or with
a screen, then the momentum of the field is transmitted to the ob
stacle. The momentum transmitted normally to unit area in unit
124 jjLECTUODYXAMrcs [Part II
time is nothing other than the pressure (since momentum trans
mitted in unit time is force). For this reason, electrodynamics pre
dicts that electromagnetic fields (and, as a particular case, light
waves) are capable of exerting a pressure on matter.
Angular momentum of a field. According to (13.27), the field
momentum density is
From this it follows that the angularmomentum density is
and the total angular momentum of the field is
M / f[r[EII]]dP. (13.28)
JrTCC J
Linear momentum and angular momentum of a field satisfy the
conservation laws together with similar quantities for the charge
contained in the field. The value of the angular momentum of a
field is very essential in the quantum theory of radiation.
Sec, 14. The Electrostatics of Point Charges.
Slowly Varying Fields
An important class of approximate solutions of electrodynamical
equations comprises slowly varying fields, for which the terms
~ ( A and ' in Maxwell's equations can be neglected. The re
maining terms form two sets of equations, which are entirely inde
pendent of each other:
(14.1)
rotE = (14.2)
and
divH = 0, (14.3)
rot II = J 1  . (14.4)
The first two equations contain only the electric field and the den
sity of the charge producing the field; the second two equations
involve only the magnetic field and current density, the righthand
sides of the equations being regarded as known, functions of coordinates
and time. Since there are no time derivatives in (14.1)(14.4),
the time dependence of the electric field is the same as the electric
Sec. 14] THE ELECTROSTATICS OF POINT CHARGES 125
charge densities, and the time dependence of the magnetic field
is the same as the current densities. Hence, to the approximation
of (14.1)(14.4), the field is, as it were, established instantaneously,
in correspondence with the charge and current distribution that
generated it.
The fact is that any change in the field is transmitted in space
with the velocity of light c. If we consider the field at a distance E
from a charge, the electromagnetic disturbance will reach it in a
r>
time . The charge, of velocity v, will be displaced, during that
c if
time, through a distance r . The approximation (14.1)(14.4)
c R
can be applied only when the displacement v does not lead to
c
any essential redistribution of the charge. For example, let a system
consist of two equal charges of opposite sign, which succeed in chang
ing places in a time . Then, the electric field at a distance 12, at
It c
the instant i = , will have a direction opposite to the one it had
during the instantaneous propagation at the instant t 0.
Hence, if the dimensions of the system of charges arc r and their
velocities v (r and v determine the orders of magnitude), then equations
(14.1)(14.4) can be used at the distance R from the system, for
r f? c
which the inequality > or R <^ r is satisfied.
We shall consider the limiting case, when v <^ c. Then the region
of applicability of our approximation will be very large.
Equations (14.1), (14.2) are called the equations of electrostatics,
and (14.3) and (14.4), the equations of magnetostatics.
Scalar potential in electrostatics. In order to satisfy equation (14.2),
we put
E = V9. (14.5)
According to (14.20), 9 is the scalar potential. The equation for the
scalar potential is obtaining from (14.1)
divgrad9 = A9 = 4jcp , (14.6)
which also follows from (12.38) if we equate to zero the nonstatic
1 02

Let us find the solution to equation (14.6) for a point charge, i.e.,
we put p equal to zero everywhere except at one point of space.
Let us put the origin at this point. Then 9 can depend only on the
distance from the origin r.
In Sec. 11 an expression for the Laplacian A was obtained in
spherical coordinates (11.46). In the special case, when the required
function depends only on r, we obtain from (11.46)
126 ELECTRODYNAMICS [Part II
I d 9 d<?
Let us integrate this equation between r l and r 2 , first multiplying
it by r 2 . Since the region of integration does not contain the origin,
where the point charge is situated, the integral of the righthand
side becomes zero. Hence,
r2 ^9 _ 2 d <* . r 2 A'. A const
/ . ' f i . , /  ^j_ Lfdiioi' .
2 dr 2 1 dr l ' dr
Therefore the potential is
The constant B is equal to zero if we take the potential to be equal
to zero at an infinite distance away from the charge. Let us now
determine the constant A. For this, we integrate equation (14.6)
over a certain sphere surrounding the origin. Since the Laplacian
A 9 is div grad 9, the volume integral can be transformed into an
integral over the surface of the sphere. This integral is
grad9rfs I % r 2 fZi= 471^4 .
On the righthaud side we have
since the integration region includes the point where the charge is
situated. Thus A  e.
The potential of the point charge is
e
9y. (14.8)
We obtain the same thing for a spherically symmetrical volume
charge distribution, if the potential is calculated outside the volume
occupied by the charge. In other words, the potential of a charged
sphere at all external points is the same as the potential of an equal
point charge situated at the centre of the sphere. A similar result
is, of course, obtained for the gravitational potential. This fact is
used in most astronomical problems, where celestial bodies are con
sidered as gravitating points.
If the origin does not coincide with the charge, and the charge
coordinates are #, y, z (radius vector r) then the potential at point
X, Y, Z (radius vector R) is
Sec. 141 THE ELECTROSTATICS OF POINT CHARGES 127
The potential of a system of charges. The potential due to several
charges e v e 2 , e 3 , . . . , e,, . . . , whose positions are given by the radius
vectors r 1 , r 2 , . . . , r 1 ', at the point R, is
9 =
~ R r' ^ V~(X ~
Using the summation convention, any radicand in this formula
can be rewritten in the form
But, in order to save space, we shall use the notation (X\ x^) 2
instead of (X\ x^) (X^ x^) . Then the potential due to a system
of point charges is written as
But we must remember that inside the brackets is a summation
for X from 1 to 3.
Note also that the potentials due to separate charges at the point
R are additive, since equation (14.6) is linear in 9. And so the full
solution, due to all the charges, is equal to the sum of all the partial
solutions for each charge separately.
The potential due to a charge system at a large distance. Let us
now assume that the origin is situated somewhere inside the region
occupied by the charges (for example, at the centre of the smallest
sphere embracing all charges), and that all the radius vectors r 1 '
satisfy the inequalities
B > r'. (14.11)
In other words, we shall look for the potential of a system of charges
at a great distance from it. Then the function (14.10) can be expanded
in a Taylor's series in terms of a;', y\ z { . We shall perform the expansion
up to the quadratic term inclusively, but we shall first write it only
for one term of the sum over all the charges, omitting the index
i. The expansion is of the form:
y)* + (Z *)]'/,
. + !*..  []'/, . (14.12)
The summation convention permits writing in concise form the Taylor
series for a function of several variables. Since Xl~ R 2 , we obtain
the expression for the first derivative
i   __   _^~   1   X n 4r 1 *\
dXv. R~~dX~v.dR R~ R*~' l 1 * 1 ^
128 ELKCTKOOYN'AMICS [Part II
where we have used equation (11.36) which, in the notation of this
section, is of the form ~ f '  ' * .
V \ \j. / 1
Thus, the term in the sum (14.12), which is linear in r [JL , is equal to
j,i A' M >X I // Y i zZ _ T R ,
/j>n yJ3 "~~ yxi " ^14, 14)
it is somewhat more difficult to calculate the term which is quadratic
in ,r {Jl , ov We first write down the second derivative:
6 2 I __ d_ Xv. _ __ 1 d_X^_ _ Y _^ __ L
iJX l Lci'Xv~/t " c'Yv 7? 3 /?J~~0~Y7 ' L 1TX^ 7? 3 '
r rhe partial derivative ' X is equal to zero for a / v and to 1 for
o JVv v
[ji = v. Further,
f; 1 c)Jf d _} _ __ A\ 3 :r.Y v
c^ A\ "A" cA\ ^7? ~Zf 3 "" " S" " ~/i T " "" '^ 5 "
}>y the rule for differentiation of involved functions.
Thus, we obtain
__^* _ i_ L dX > Ji S.YnXv
"^X^~A\ "77 /e 3 ^ a A^ v "^ ^/e 5 "" "
Hence, the required expansion ;R rj 1 is of the form
We shall now subtract from the quadratic term a quantity equal
to zero:
r 2 1 c* 1 __ .r  //a i : 2 l^
2 ' , 4 i r'A'j" "" " (5 "^~/f
[A ^   from (14.8)]. Then the last term in (14.15), written in terms
dvanta
r _
~ '
of components, is (taking advantage of the fact that ^ v  ^v
' ( T) + **( '" )( J  ;*}(*2*  id.
Here, it is quite obvious that a term equal to zero has been sub
tracted, for A" 2 + Y*+Z 2 =^R 2 . Rearranging the terms, we have
_ i \ , L 2 rM
IP i + r r) x
3A '^ , . srzi
** ___ _L_ "^t/y __ I
M 7? 5 ^^ 2 7?5 J
See. 14]
THE ELECTROSTATICS OF POINT CHARGES
129
The expansion (14.15) must be substituted in the equation for
potential (14.10) and summed over all the charges. We introduce
the following abbreviated notation:
d^ y>,r<; (14.16)
(14.17)
(14.18)
The vector d (i.e., the three quantities d x , d y , d z ) and the six
quantities q xx , q YY , qzz, qxy, qxz, qyz, depend only on the charge
distribution in the system, and not on the place at which the potential
is determined. In the notation of (14.16)(14.18), the potential
at large distances away from the system is of tho form
^ __ x, , (dR) , 1
with the terms of different indices of the type q xy actually appearing
twice in the summation (for example, q 12 and the equal term r/ 21 ).
The vector d is called the dipole moment of the charge system.
The six quantities q are called the quadrupole moment components.
The dipole moment. We shall now examine the expression obtained
for potential. The zero term
corresponds to the approximation
R
according to which all the charge is considered to be concentrated
at the origin. In other words, it corresponds to a substitution of the
entire charge system by a single point charge. This approximation
is clearly insufficient when the system is neutral, i.e., if JT' e t  = 0.
This case is very usual, since atoms and molecules are neutral (their
electronic charge balances the charge on the nuclei).
90060
130 ELECTRODYNAMICS [Part II
Let us assume that the total charge is equal to zero and then
consider the first term of the expansion, involving the dipole moment.
This term decreases like^ r , i.e., more rapidly than the potential
of the charged system. Besides, it is proportional to the cosine of
the angle between d and R. The simplest thing is to produce a neutral
system by taking two equal and opposite charges. Such a system
is called a dipole. Its moment is
d = elF = e(r l  r 2 ) (14.20)
in accordance with the definition (used in general courses of physics)
that the dipole moment is the product of charge by the vector joining
the positive and negative charges.
It can be seen from equation (14.20) that the definition of dipole
moment does not depend on the choice of coordinate origin, since
it involves only the relative position of charges. We shall show that
the dipole moment always possesses this property.
Indeed, if we displace the origin through some distance a, then
the radius vectors of all the charges change thus:
r 1 = r' 1 + a
Substituting this in the expression for the dipole moment, we obtain
because J^ej = 0.
But if the system is not neutral, then we choose a in the following
manner :
27*
This choice is analogous to the choice of centre of mass for a system
of masses. Thus, we can say that the vector a determines the electri
cal centre of a system of charges. For a neutral system it is impos
sible to determine a, since the denominator of (14.22) is equal to
zero. If for a charged system we choose a according to (14.22), then
J^ir'' = 0, i.e., the dipole moment of a charged system, relative
t
to its electric centre, is equal to zero.
We thus have the following alternatives : either the system is neutral,
and then the expansion (14.29) begins with a dipole term independent
Sec. 14] THE ELECTROSTATICS OP POINT CHARGES 131
of the choice of coordinate origin, or the system is charged, and then
the dipole term in the expansion is equal to zero for a corresponding
choice of origin.
Quadrupole moment. In the expansion (14.19), we now consider
the second term containing the quadrupole moment. A quadrupole
is a system of two dipoles of moment d, which are equal in magnitude
and opposite in direction. It is clear that a potential expansion for
such a system will have neither a zero nor first term, so that equation
(14.19) contains only a second term on the righthand side. The sim
plest quadrupole can be formed by placing four charges at the vertices
of a parallelogram, where the charges are of equal magnitude but
with pairs of charges having opposite signs. The charges alternate
when we traverse the vertices of the parallelogram. Such a system
is neutral. However, a charged system, too, can have a quadrupole
moment. It indicates to what extent the charge distribution in the
system differs from spherical symmetry. Indeed, in this section it
was shown that the potential due to a spherically symmetrical charge
system decreases in strict accordance with a = law, and the potential
due to a quadrupole follows a ^ 3  law. For this reason, the quadrupole
term in the potential expansion can arise only in the case of a non
spherical charge distribution.
The principal axes of a quadrupole moment. Let us now determine
in what sense the quadrupole moment characterizes a nonspherical
distribution. In equation (14.22), an analogy was established between
the centre of inertia of a mass system and the electric centre of a
system of charges. In a similar way, equations (14.17), (14.18) allow
us to establish a certain correspondence between the components
of quadrupole moment and moments of inertia of a system of masses
Jxx, . . ., Jyz (see Sec. 9), defined in equations (9.3).
We can disregard the fact that a summation appears in equations
(14.17) and (14.18), while (9.3) involves integration. This difference
will not exist if we take a continuous charge distribution or a dis
crete mass distribution (as for nuclei in a molecule). We shall take
the latter. In addition, we shall forget for a moment that the compo
nents of the moment of inertia involve masses and not charges.
Then the relationship between the quadrupole moment and the
moment of inertia is of the form:
Qxx ^ Jxx + 3" (Jxx + Jyy + Jzz), Qxy ^ Jxy,
(Jyy ~ Jyy + ~ (Jxx + Jyy + Jzz), <fxz ^ ~ Jxz,
Qzz ^ Jzz + q (Jxx + Jyy + Jzz), Qyz ^ Jyz
132 ELECTRODYNAMICS [Part II
The sign ^ above the equality symbol indicates correspondence
between terms. Indeed, according to (9.3) the first line gives
 r '' 2 ) =
i i
The relations in the second column are obvious.
In Sec. 9 it was shown that moments of inertia can be reduced
to principal axes, i.e., a coordinate system can be found for which
the products of inertia are zero. But since the relations between q
and J are true for any coordinate system, the components of the
quadrupole moment of different signs also become zero in these
same principal axes. In the principal axes, the quadrupole moment
is expressed, in terms of moments of inertia, as
(14.23)
If the system possesses spherical symmetry, then J 1 r=J 2 ^J 3 ,
so that q l ~q^q 9 ^o. Therefore, the presence of a quadrupole
moment in a system of charges indicates that the charge distribution
is not spherically symmetrical. However, a reverse assertion would
not be true: if the quadrupole moment is equal to zero, the system
of charges may not be spherically symmetrical. It will then be neces
sary, in expansion (14.19), to take into account terms of higher order.
It will be noted that from (14.23) there follows directly the identity
7i ~t #2 ~l ft 0, i e > on ly t wo f fche three principal components of
a quadrupole moment are independent.
The relations (14.23) should be regarded literally if we are talking
about a gravitational potential. We know that the earth is not strictly
spherical, but is flattened at the poles. Therefore, the earth's gravi
tational force contains terms which are not governed by an inverse
square law. This affects the motion of the moon, and ah 1 the more
so that of artificial satellites moving closer to the earth.
Quadrupole moment when axes of symmetry exist. Equations (14.23)
become simpler if two moments of inertia of a body are equal, for
example, J l ^J. 2 . Then
ffi^y^.^i)^  I,
ft ^ y (^3  ^i)   y >
2
Sec. 14] THE ELECTROSTATICS OF POINT CHARGES 133
In this case, the quadrupole moment has only one independent
component q. Its sign is called the sign of the quadrupole moment.
The quantity q = ei (z? ^1 .
If the charges were distributed with spherical symmetry, we
would have the equality JT r eif 4<l = 3 JT'efZ 1 '*, for then e < xi * ~
i i i
2^c { y { ' eiZ* and q, too, would be equal to zero.
The positive sign of q shows thatJTV/z' 2 > ^ J^r*", i.e., it indicates
i i
a charge distribution extending along the caxis. From (14.19), the
potential due to such a quadrupole with one component q is
1 /3.Y 2 1 \ 1 /33' 2 1 \ q 13Z* 1 \ __
4 ? \7fT  & }  4 9 (fir  Jp) 1 2 \ R r ~&] "
*'' r 8 2Z 2 \ 3 /#; 3^
/.^ / " 4 ? \ v? 6
Thus, the potential of a quadrupole depends on the angle & according
to the law ]  3 cos 2 &, where & is the angle between the axis of
symmetry of the quadrupole and the radius vector of the point
at which the potential is determined.
Similar deviations from spherical symmetry have been found in
the electrostatic potential of many nuclei. The quadrupole moments
of nuclei give us an insight into their structure.
The energy of a system of charges in an electrostatic field. We
shall now calculate the energy of a system of charges in an external
electric field. The potential energy of a charge in a field is equal
to f/pcp, because the force acting on the charge is equal to
F   Vt/ eV9^eE. The energy of a system of charges is thus
17= 5>, 9 (r'), (14.25)
where r 1 ' is the radius vector for the ith charge.
Let us suppose that the field does not change much over the space
occupied by the charges, so that the potential at the site of the ith
charge can be expanded in a Taylor's series:
9  * (0) + *+ *^ . (14.26)
134 ELECTRODYNAMICS [Part II
We shall transform the last term in the same way as in the expansion
(14.15); taking advantage of the fact that 9 is the potential of the
external field (and not the field produced by the given charges),
so that A?=0> we subtract from 9 the quantity 3 A? equal to zero.
Then, after summation over i, we obtain
(U.27)
Here, the value of the field (V9) = E at the origin has been
substituted into the term involving the dipole moment. Relating
equation (14.27) to the principal axes of the quadrupole moment, we get
 (dB.)  7l + q z + ?, ) . (14.28)
111 the case of a neutral system, the term involving dipole moment
is especially important. The quadrupole term accounts for the ex
tension of the system, since it involves field derivatives. It is interesting
to note that if the system is spherically symmetrical, i.e., if it has
a quadrupolo moment equal to zero, there is no correction to finite
dimensions. Higher order corrections are also absent, so that the
potential energy will always depend only on the value of the potential
at the centre. This is why spherical bodies not only attract, but
are also attracted, as points. Of course, these assertions are mutually
related by Newton's Third Law which holds for electrostatics, since
the field is determined by the instantaneous configuration of charges.
Exercises
1) Show that the mean value of the potential over a spherical surface is
equal to its value at the centre of the sphere, if the equation A 9 = is satisfied
over the whole volume of the sphere. Relate this to the result obtained for
the potential energy of a spherically symmetrical system of charges in an
external field.
The potential should be expanded in a series involving the radius powers
of the sphere. In integration over the surface, all the terms containing x, y,
and z an odd number of times become zero. The terms containing x t y t and z
an even number of times can be rearranged so that they are proportional
to A 9, A A 9 , and so on. There remains only the zero term of the expansion,
which proves the theorem.
2) Calculate the electric field of a dipole.
Sec. 15] THE MAGNETOSTATICS OF POINT CHABQES 135
Sec. 15. The Magnetostatics of Point Charges
The equations ol magnetostatics. In the previous section it was shown
that if the velocities of the charges are small in comparison with
the velocity of light, then the magnetic field satisfies the following
system of equations:
divH = 0, (15.1)
rotfl = j. (15.2)
c
They are called the equations of magnetostatics. From equation
(15.2) it follows that
4rc  1 ' a =0. (15.3)
Thus, for (15.2) to make strict sense, the currents must bo subjected
to the condition div j 0. But this condition is not directly satisfied
for point charges, and only the charge conservation equation (12.18)
holds.
Mean values. The condition div j = for moving point charges
can only be satisfied for an average over some interval of time t Q .
We shall define the mean value of a certain function of coordinates
and velocities of charges / (r, v) in the following way :
7=f f/(r,vM<. (15.4)
*0 J
This averaging operation is commutative with differentiation of
the function with respect to coordinates, since it is performed with
respect to a fixed frame of reference and not to the coordinates of
the charges.
Let us now average equation (12.28):
Integration of the righthand side part can be performed directly:
Let us now assume that the difference p (t Q ) p (0) increases
more slowly than the time interval t Q itself. Then, if we choose
sufficiently large, the ratio p '""" p ' ' can be as small as required.
*o
Because of this, the mean value of the current may indeed satisfy
the equation
136 ELECTRODYNAMICS [Part II
divj0. (15.5)
Consequently, equation (15.2) and all subsequent equations in
this section must be regarded as mean with respect to time; this
will be denoted by a bar over each quantity relating to the motion
of charges (no bar will be put over II).
The definition of steady motion. Let us assume that the condition
lim ' ~"  : is satisfied not only for charge density, but also for
/ ft >oo *0
any function relating to the motion of charges. Such motion is termed
stationary or steady.
A special case of stationary motion is periodic motion, for example,
cyclic motion. However, for a stationary state it is sufficient that the
charges remain all the time in a limited region of space, for the
difference / ( )  / (0) then remains finite.
The equations of this section will relate to the stationary motion
of point charges.
The equations for vector potential. In order to satisfy equation (15. 1 ),
we put, as in Sec. 12 [see (12.8)],
II rot A, (15.6)
where A is the vector potential. Equation (15.6) does not fully deter
mine A since, if we add to A the gradient of any arbitrary function /,
as in (12.30), the expression for II will not change. Thus, an additional
condition must be imposed on A. The Lorentz condition suggests
that we must have
divA0. (15.7)
Then, substituting (15.6) in (15.2), we obtain
rot II = rot rot A   . (15.8)
But according to (11.42)
rot rot A = grad div A A A =  A A , (15.9)
and we have used condition (15.7). Therefore, A satisfies the equation
AA=^j = ^Lpy, (15.10)
which is entirely analogous to equation (14.6) for the scalar potential.
Equation (15.10) can be obtained from (12.37) directly if we discard
the term 3  5 which is superfluous in magnetostatics.
The vector potential lor a point charge. The solution of (15.10)
appears exactly the same as the solution of (14.6) given by equation
(14.8) for a separate point charge: each component of A satisfies
Sec. 15] THE MAGNETOSTATICS OF POINT CHARGES 137
equation (14.6), the only difference being that on the righthand
side there appear the functions ^ ov*, ^ p^ y , * pv*. it follows
c c c
that the vector potential for a point charge is
We shall now show that A satisfies condition (15.7). The divergence
must be taken with respect to the radius vector R at the point at
which A is determined.
But V R  R r  = V r ] R r 1 , so that
,. 4 e ._ 1 e rT I e d
di v A =  W R ~ _  vV r ^777 ' ' * T IT
(15.12)
Hie expression on the righthand side of this equation is the total
time derivative of the quantity p^  rrr . From the steady state
j K r () 
condition, it is equal to zero.
The BiotSavart law. We shall now calculate the magnetic field
of a point charge. Using equation (11.28), we obtain
rl
This equation refers, of course, only to steady motion. In particular
it is applied to constant current.
Vector potential at large distances from a system of stationary
currents. The vector potential for a system of point charges is equal
to the sum of vector potentials for each charge separately:
We shall now obtain approximate formulae which are valid at large
distances from the system, similar to those obtained in electrostatics.
For this, we substitute an expansion of inverse distance in (15.13)
[see (14.14)].
1  l  r ' R (15 U]
Rr'', ~ R >"' < 1& ' 14 >
The quadratic term is not taken into account this time. The ex
pression for vector potential, to the approximation of (15.14), is
of the form
138 ELECTRODYNAMICS [Part II
since f 1 = v 1 .
The zero term of the expansion is a total derivative and, by the
steadystate condition, is equal to zero. We now transform the first
term of the expansion, using the identity
= tZe* (RrO r' = (Rr<) v' +2> (BV) r< . (15.16)
i
From this identity it follows that in (15.15) we can substitute
half the difference of the expressions on the righthand side of (15.16).
Then the vector potential will be
A = T
We now interchange the signs of the summation and vector product :
Magnetic moment. The sum appearing inside the brackets in (15.18)
is called the magnetic moment of the system of charges (or system
of currents). The mean magnetic moment is written thus:
The equation for the vector potential (15.18) can be written, by
means of the magnetic moment, in the following form:
< 15  20 >
The field ot a magnetic dipole. Let us now calculate the magnetic
field. By definition
H=rotA=rot[v~ ,jji].
Since jl is a constant vector, equation (11.30) gives
H
Sec. 15] THE MAONETOSTATICS OF POINT CHARGES 139
because A~p = Further, (ft V) R=jZ [see (11.35)]. In order to cal
culate (f*V)jjo, we use equation (11.36). This yields
Finally, collecting both terms, we arrive at an equation for H:
H = 8B(Bfl
For comparison, we deduce the expression for the electric field
of a dipole:
(15.22)
Thus, both expressions for the field (both electric and magnetic)
are of entirely analogous form. The only difference is that, instead
of the electric moment, the equation for magnetic field involves
the magnetic moment. This explains its name.
In the case of a charge moving in a flat closed orbit, the definition
of magnetic moment (15.19) coincides with the elementary definition
of moment in terms of "magnetic shell." As was shown in Sec. 5
[see (5.2), (5.4)], the product [rv] is twice the area swept out by the
radius vector of the charge in unit time, or [rv] = 2 y . By definition
of the mean value (15.4)
ji = J fl^ldJ = 4s. (15.23)
^ t J c dt ct v '
o
Here, t Q is the time of orbital revolution of the charge. In this time,
the charge passes every point on the orbit once; hence the mean
current is equal to / = . This yields the definition of magnetic
moment familiar from general physics:
ji = ^ . (15.24)
The similarity of equations (15.21) and (15.22) shows the equivalence
of a closed current (i.e., a magnetic shell) and a fictitious dipole
with the same moment p. The field at large distances from a system
of currents is produced, as it were, by the effective dipole.
The relationship between magnetic and mechanical moments. An
especially interesting case is that when all the charges of a system
are of the same kind (for example, when they are all electrons).
140 KLECTRO DYNAMICS [Part II
Then the magnetic moment is proportional to the mechanical moment.
Indeed, for a system of charges with identical ratios , we obtain
Pi   rt y [r'Vl  b  Fm [r' VJ =  V[r'p']    M .
r* 2c 4* l J 2mc^*t l J 2mc 4* L rj 2mc
(15.25)
Equation (15.25) has very important applications.
A system of point charges in an external magnetic field. We now
consider the question of the interaction of a system of currents with
an external magnetic field. For this we must have an equation
describing the interaction of a point charge with the field. We obtained
equation (13.17) for the general spatial charge distribution. In this
equation, the transition to point charges is obtained by changing
the integral to a summation over the charges. The term obtained
for action is of the form
where the indices i in A and cp denote that the potentials are taken
at the same point as the ith charge.
In magnetostatics, only slowly moving charges for which v<^.c
are studied. Newtonian mechanics can then be applied to their motion.
In the absence of a field, the action function of the particles is of the
form
It will be shown in Sec. 21 that this expression holds only when r< c.
In magnetostatics, where this condition is satisfied, the action function
of a system of charges in an external field is obtained by adding
(15.27) and (15.26):
The field due to the charges themselves does not appear in this
equation. The expression for the integrand is the Lagrangian of the
system. It involves velocity linearly as well as quadratically (in
the expression for kinetic energy), and, for this reason, does not have
the form that we used in Part I, L = T U.
However, the general relationships still hold. Therefore, from the
Lagrangian the expression for momentum is obtained in terms of
velocity
P'^^' + ^ (15  29)
Sec. 15] THE MAGNETOSTATICS OF POINT CHARGES 141
Let us determine the energy in terms of momentum using the basic
equation (4.4):
A,V    A..V + <,
< 15  30 >
so that the term which is linear with respect to velocity is eliminated
from the expression for energy in terms of velocity.
The Ha mil to ni an function for a system of charges in an external
magnetic field. The linear term in velocity of the Lagrangian affects
the form of the Hamiltonian. Let us write down 3? from its definition
(10.15). To do this it is necessary to substitute into the energy ex
pression, by means of equation (15.29), momenta instead of velocities :
v 1 '  (?,   A,) . ( i (15.31)
mi \ FI c V \\ ^ v '
The Hamiltoman is ; I V 1 l
'' (15 ' 32)
Let us assume that the magnetic field, in which the system is
situated, is weak and uniform (at least within the limits of the system).
The vector potential for a homogeneous field will be represented as
A = *[Hr] . (15.33)
Indeed, then rot AH [from (11.30), (11.35) and (11.33)]. And also
div A0 from (11.29) and (11.34).
Since the magnetic field is weak we can neglect in (15.32) the term
involving A?. Then, substituting (15.33) in (15.32), we find an ex
pression for J^:
< 15  34 >
The last term in (15.34) gives the required addition to the
Hamiltonian. Since this term is proportional to H, we can replace p,
by w,v in it to the same accuracy, i.e., neglecting terms of order H 2 .
Performing a cyclic permutation of the factors in (15.34) and putting
the sum inside a vector product sign, we obtain an expression for
the addition to the Hamiltonian:
X" =  (H JT^ [rVj) =  (Hn) . (15.35)
i
This expression is very similar to the energy of a system of charges
in a homogeneous electric field which involves only the electric dipole
moment of the system of charges. Note that this term is of the form
142 ELECTRODYNAMICS [Part II
(dE) [see (14.28)]. This indicates a further similarity between
electric and magnetic moments.
Larmor's theorem. Let us now compare the expression for the
momentum of a charge placed in a constant homogeneous magnetic
field with that for the momentum of a particle relative to a rotating
coordinate system. From (15.29) and (15.33), the former is
 [Hr] 9 (15>36)
the latter can be easily found from (8.5):
p = mv' + m[cor]. (15.37)
We now consider a steadily moving system of identical charges
(for example, an atom or molecule); the nuclei, being heavier, are
regarded as fixed. Let us assume that in the absence of any external
magnetic field, the motion in the system is known. Then, comparing
equations (15.36) and (15.37), it is easy to see that if we consider the
motion of these charges relative to axes rotating with angular velocity
<>=, (15.38)
it will not differ from motion relative to fixed axes in the absence
of a magnetic field. The equations of motion relative to rotating
axes will have their usual form p,=F,, where F, is the force acting
on the ith charge in the absence of any external magnetic field,
because the correction to the momentum due to the angular velocity G>
[defined in accordance with (15.38)] will cancel with the correction
due to the magnetic field. The magnetic field must be sufficiently
weak so that the change in magnetic force with rotation can be
neglected.
We can say, therefore, that, with the application of a constant
and uniform weak external magnetic field, a system of charges with
identical ratios begins to rotate with a constant angular velocity
_ //
 WL  = ~2~  . This statement is called Larmor's theorem, and G> Z
is the Larmor frequency.
Precession of the magnetic moment. If a system possesses a magnetic
moment fx for motion which is undisturbed by a magnetic field,
then, when a magnetic field is superimposed, this moment will move
around the direction of the magnetic field, similar to the free top
in equations (9.14) (note that <o is in the direction of the axis of
rotation, i.e., in the direction of the magnetic field). The precession
of magnetic moment about the field is called the Larmor precession.
Magnetic moment in an inhomogeneous field. Let us suppose that
the magnetic field possesses a small inhomogeneity. Then, in the
equations of motion, the term
Sec. 15] THE MAGNETOSTATICS OF POINT CHARGES 143
F =  VJf ' = V (Hfji) (15.39)
denotes the force acting on the moment and tending to move it
as a whole. Expanding (15.39) by (11.32), we obtain
But for an external field rot H is equal to zero so that the force is
(15.40)
This is the wellknown force of attraction to a magnet. It is
maximum near the poles of the magnet where the inhomogeneity
of the field is greatest.
Exercise
Study the magnetic moment y. moving in a magnetic field given by the
components H z = H ; H x = H cos <o t; H y = H l sin w t. Consider the
e #o *
cases co = to ~o~~"~ <*> > 0.
The equation describing a vector rotating with angular velocity o/, is,
according to Sees. 8 and 9,
Whence we obtain an equation for the precession of magnetic moment in a
magnetic field
By multiplying both sides of this equation by n, we convince ourselves
that the absolute value of magnetic moment jx is conserved. It is, therefore,
sufficient to write the equations only for the components \i x and (jt y , replacing
Vz by V V? p \f y .
Using the abbreviated notation to == j co t = [Cf. (1.5.38)], we
multiply the equation for (i y by * and combine with the equation for
to obtain
(Vx ipy) = ivodtx *^y) tttjeiwl V JX 2 I** I*
We seek the solution in the form y. x *Vy ^ e ^ f ^ , and get the
following equation for amplitudes A :
Multiplying these equations, we find
A A V 2 ! A l 1 ^
A + A  = ^  ^ 2 ; A = . i
When 6> = eo ("paramagnetic resonance"), the moment rotates in the
plane x y with frequency o . When <> > 0, i.e., in the case of an infinitely
slow rotation of the field, the moment strictly follows the field, its direction
all the time being the same as that of the field.
144 ELECTRODYNAMICS [Part I]
Sec. 16. Electrodynamics of Material Media
Field in a medium. We know that material media consist of nuclei
and electrons, i.e., of very small charges in very rapid motion. There
fore, in a small region of a body a region having atomic dimensions
all electromagnetic quantities (field, charge density, and current;
change very rapidly with time. In two neighbouring small regions
these quantities may, at the very same instant, have completely
different values. Therefore, if we examine the field in a medium ful]
of charges in detail, then we will observe only a rapidly and irregularly
varying function of coordinates and time.
Mean values. The inhomogeneities of a field are of atomic dimensions,
However, such a detailed picture of the field is not usually of an}
interest. As usual, in any description of macroscopic bodies, it is
essential to know mean values for a large number of atoms. For exam
ple, in mechanics, mean density values are used. For this mean tc
have any significance, we must isolate a volume of the body contain
ing a large number of atoms, determine its mass and divide by the
volume.
This volume must be so large that the microscopic atomic structure
of the substance cannot affect the mean value of the density. At the same
time, the mean macroscopic value must be constant over that volume,
This will be readily seen from the following. Let the volume be arbi
trarily divided into two equal parts. Then the mean for each part
should not differ from the mean for the whole volume.
Such a volume is termed physically infinitesimal. We shall call it
F . If we take all its dimensions to be large compared with atomic
dimensions, then the mean value should not depend on the shape oi
the surface bounding the volume; the latter may be spherical, cubi
cal, etc.
Besides averaging over volume, it is also necessary to perform
averaging over time. The interval of time over which the average i&
to be taken must be large compared with the times of atomic motions,
though still sufficiently small so that the mean values over two semi
intervals do not differ from one another.
Let the volume have the form of a cube of side a. We shall denote
the coordinates of its centre by x, y, z. The time interval, over which
the averaging is performed, will be called , and the instant corre
sponding to the centre of the interval will be denoted by t. The coordi
nates of any point inside the cube, relative to the centre, will be called
, TJ, , and instants of time measured from t, will be denoted by &.
Thus, the limits of variation of the quantities are given by the follow
ing inequalities:
Sec. 16] ELECTRODYNAMICS OF MATERIAL MEDIA 145
The actual value of any quantity at a definite instant of time is
/ (x + %, y+f\, z + , *+&). It is related to a mathematically infinitely
small volume dV^dZ, df\ d^ and an interval of time db. The average
value, over a physically infinitesimal volume V and interval of time
, is obtained if / is integrated over dV dt and the integral divided
by F , in accordance with the usual definition of an average :
J(x, y, z, t) 
a / a l a l V
(16.1
This average value / (x, y, 2, t) refers to the point x, y, z, and time t.
The electrodynamics of such mean values is termed macroscopic, as
opposed to microscopic, which has to do with a field due to separate
point charges and a field in free space.
The mean thus determined is differeiitiable with respect to time and
coordinates. As parameters it involves the coordinates of the centre of
a physically infinitely small volume x, y, z, and the time t. Obviously,
we can differentiate with respect to these values:
^7(*,y, M) 
a /2  /, /, '/
^~VT S d *> J d ^ S d ^ J^^/^iS' yv,*\l, t+*)
i0 /. /. /. v. _ dj
_,_. (lb . Z)
In other words, the mean value of the derivative of a quantity is
equal to the derivative of its mean value.
Density of charge and current in a medium. Under the action of
the electric and magnetic field, there occurs a redistribution of the
charges and currents in any substance. When Maxwell's equations
are averaged, the mean density of the redistributed charge is p and
that of the current is j. We shall express p and j in terms of other
values which will later make it possible to give the averaged Maxwell
equations a very symmetrical form.
We define the dipolemoment density in a substance by the follow
ing formula:
(16.3)
The dipole moment P in unit volume is called the electric polarization
of the medium. If the substance is completely neutral, its dipole
moment is uniquely determined as ^e^r 1 ' [see (14.21)]. Going over
to a continuous charge distribution, we write
100060
146 ELECTRODYNAMICS [Part II
d=fpreZF. (16.4)
Integral (16.3) can be identically written in the form:
JPdF = JrdivPdF. (16.5)
This relationship can most simply be proved by writing in terms of
components, for example,
(xP x
Xl
V % **
f \x(P y ) dxdz+ j (x(P z ) dxdy J J fp*da;di/dz.
The limits of integration are at the external boundaries of the medium,
where the values P*, P y , P* are zero. This proves (16.5). Comparing
(16.4) and (16.5) we obtain
J r (div P + ~p) dV = . (16.6)
However, since the shape and dimensions of the body are arbitrary,
the quantity
divP + p = (16.7)
must be zero.
Thus, the mean density of a charge "induced" by the field is equal
to the divergence of the electric polarization vector taken with oppo
site sign.
In a similar way, we can express the mean density of induced cur
rent. To do this we define the magnetic polarization vector, equal to
the magneticmoment density, as
(16.8)
But the magnetic moment, by definition, is expressed as
_ ,.j Q. fr 1 ' v'1
= 2^ 2 . Applied to the current distribution, this gives
We shall now prove an identity analogues to (16.5):
(16.10)
Sec. 16] ELECTRODYNAMICS OP MATERIAL MEDIA 147
For this, it is simplest to go over to components:
[rrotM]*dF = J (yrot*Mzrot y M)dF =
The terms y and z ~ are integrated by parts. All
the integrated quantities become zero when the limits are inserted,
so that, in agreement with (16.10), only 2 M x remains. Now, comparing
(16.9) and (16.10) we obtain
.dF = j^^dV . (16.11)
In order to determine j fully, we calculate its divergence and apply
the charge conservation law (12.18) written for mean values [see
(16.2)]:
div j   ^  div ^ . (16.12)
From (16.11) and (16.12), j is uniquely determined as
j^^ + crotM. (16.13)
Indeed, this expression satisfies both equations. Finding the diver
gence of both parts of (16.13), we arrive at (16.12), since div rot M = 0.
8 If
Further, substituting the quantity^ in the lefthand side of (16.10),
we get
According to (16.3) and (16.4), P is replaced by pr. But, [r, pr] = 0,
OT>
so that the term ^ does not contribute to equation (16.11). The
identity (16.13) is thus proved.
Averaging Maxwell's equations. We shall now consider the averaging
of Maxwell's equations. From (16.2), differentiation and averaging
are commutative, so that a bar can simply be put over the first pair
in order to denote that they have been averaged:
, (16.15)
divH= 0. (16.16)
The mean value of an electric field E is called the electric field in a
medium. We shall hereafter write it without the bar, which denotes
that it has been averaged, taking it for granted that only mean values
148 ELECTRODYNAMICS [Part II
will always be taken in a medium. The mean value of the magnetic
field is called the magnetic induction and is denoted by the letter B.
ft is all the more unnecessary to write a bar over it because the concept
of induction, which is not equal to field, makes sense only for a medium.
The asymmetry in the terminology for electric and magnetic fields
will bo explained later.
In this notation, the first pair of Maxwell's equations takes the
following form:
wtE =!., (1617)
divB0. (16.18)
Now let us average the second pan* of Maxwell's equations :
(16.20)
We substitute p and j from (16.7) and (16.13) and rearrange the
terms somewhat, obtaining the two following equations (the bars are
again omitted):
rot(B 47cM)  (E + 4:rP) , (16.21)
div (E + 47rP) = . (16.22)
We introduce the following new designations:
E [47cPD. (16.23)
I) is called the electric induction.
Further,
B 47cM^H. (16.24)
II is called the magnetic field in a medium, which, therefore, does not
equal the mean value of the magnetic field in a vacuum.
In the notations (16.23) and ( 1 6.24), the second pah* is written similar
to t he first pair :
rot II ^ J~?, (16.25)
divD0. (16.26)
The similarity between (16.26) and (16.18) explains why it was
convenient to call the mean value of the magnetic field the magnetic
induction: here, both electric and magnetic induction vectors have no
sources in the medium. The similarity between (16.25) and (16.17)
justifies the term magnetic field given to the vector H (16.24).
The incompleteness of the system of equations in a medium. Thus,
due to a suitable system of notation, the first and second pairs of
Sec. 16] ELECTRODYNAMICS OF MATERIAL MEDIA 149
Maxwell's equations in a medium have, as it were, become more symmet
rical than those in a vacuum. But we must not forget that this system
has now ceased to be complete: as before, there are eight equations
(of which only six are independent) and twelve unknowns B, E, D, H
(with three components for each vector). Consequently, the system
(16.17), (16.18), (16.25), and (16.26) cannot be solved until a relationship
is found between inductions and fields. This relationship cannot be
obtained without knowing the specific structure of the material
medium.
Dielectrics and conductors. We shall consider, first of all, how charges
behave in a medium in the presence of a constant electric field. The
field will displace the positive charges in one direction, arid the nega
tive ones in another. As a result, polarization P will arise. Two essen
tially different cases can occur here.
1) Under the action of the field inside the body, a certain finite
polarization P, dependent on the field, is established. This polariza
tion may be represented vividly (though in very simplified fashion!)
as a displacement of charges from the equilibrium positions which they
occupied in the absence of the field to new equilibrium positions
much like the way a load suspended on a spring is displaced in a gravi
tational field. If a finite polarization (dependent on the field inside the
body) is established, that body is called a nonconductor or dielectric.
2) In a constant electric field acting inside the body, the charges do
not arrive at equilibrium, and a definite rate of polarization increase,
^ , is established. In this case, through every section of the conductor
dP
perpendicular to the vector ^~ , there pass electric charges or, what
amounts to the same thing, a flow of current. From equation (16.13),
the derivative TT may indeed be interpreted as a current component.
ct
As regards the second current component, c rot M, it relates to the
instantaneous value of quantities and cannot characterize the change
of anything with time. For this reason, the classification of bodies into
conductors and dielectrics is obtained from the behaviour of the
... dP
quantity ^ .
The displacement of charges under the action of a field can roughly
be likened to a load falling in a viscous medium with friction, when, as
is known, a definite speed of fall is established.
A medium, in which a constant electric field produces a constant
electric current, is called a conductor.
If a constant electric field is produced in free space and a conducting
body of finite dimensions is introduced into it (for example, a conduct
ing sphere or ellipsoid), the charges in the body will be displaced so
that a field equal to zero will be established inside the conductor. For
this, the mean charge density inside the body must also equal zero,
160 ELECTRODYNAMICS [Part II
because the lines of force of the field originate and terminate at charges.
Under the action of such a field, the charges inside the conductor
would be displaced. This means that an equilibrium will be established
inside the conductor only when all the induced charges emerge to the
surface. They will be distributed on the surface of the body so that the
mean field inside the conductor is zero, and the lines of force outside
the conductor will arrive normal to every point of the surface.
Continuous current in a conductor. A continuous current can
flow in a conductor only along a closed conducting circuit. And the
electric field must always have a component along the direction
of the circuit. Then charges of given sign will always move in one
direction, thereby producing a closed current. The work performed
by unit charge in moving around the circuit is called the e.m.f. acting
in the circuit:
.f. = J Edl. (16.27)
e. m
This formula differs from (12.1) in that E denotes the field acting
inside the conductor.
External sources ol e.m.f. In a conductor, a constant e.m.f. can
only exist at the expense of some external source of energy, for example,
a primary cell. When a current passes in the circuit of the cell, ions
are neutralized on the electrodes, thus yielding the source of energy
that maintains the e.m.f.
If, as usual, we put E = ^9, then the expression for e.m.f.
will be
.f. =  J
e.m
Therefore, the e.ra.f. may be defined as the change in potential
in going round a closed path. Thus, the potential is not a unique
function of a point : for each traverse, it changes by the value of the
e.m.f. in the circuit.
The magnetic properties ol bodies. We shall now consider the
magnetic properties of bodies. In a constant magnetic field, a definite
equilibrium state will always be established in the medium. Here
we must distinguish between the following two cases.
1) In the absence of a field, the atoms or molecules of a substance
possess certain characteristic magnetic moments that differ from
zero.
As was shown in the previous section, the energy of every separate
elementary magnet in a magnetic field is [i H. Hence, the energy
of elementary magnets that have a positive moment projection
on the field is less than the energy of elementary magnets with a
Sec. 16] ELECTRODYNAMICS OF MATERIAL MEDIA 151
negative moment projection. Atoms and molecules are in random
thermal motion. As a result of this motion and of the action of a
magnetic field, an advantageous energy state is established in which
positive moment projections on the field predominate. For more
detail about this equilibrium see Part IV.
It will be noted that the projection of an isolated magnetic moment
on the magnetic field is constant it merely performs a Larmor
precession around the field. But the interaction between molecules
disturbs the motion of separate moments, and results in the establish
ment of a state with a mean magnetic polarization other than zero.
2) The atoms or molecules of a body do not possess their own
magnetic moments in the absence of a field.
As was shown in the previous section, when an external magnetic
field is applied, the motion of charges in atoms or molecules changes
due to the Larmor precession. Indeed, a precession of angular veloc
ity co = is superimposed on motion undisturbed by a magnetic
field. In exercise 4 of this section it will be shown that this precession
leads to the appearance of a magnetic moment in a system of charges.
We shall only note here that the direction of the magnetic moment
induced by the field must be in opposition to the direction of the
magnetic field; this follows from Lenz's induction law. Indeed, an
induced current produces a magnetic field in a direction opposite
to that of the inducing field.
A substance in which an external magnetic field produces a resultant
moment in the same direction, is called paramagnetic. If the magnet
ization is in the opposite direction to the field, the substance is dia
magnetic.
Ferromagnetism. There are crystalline bodies in which the magnetic
moments are aligned spontaneously, i.e., in the absence of any external
magnetic field. Such bodies are called ferromagnetic. The magnetic
polarization of the body itself is related to the directions of the
crystalline axes. For example, in iron, whose crystals have cubic
symmetry, the intrinsic magnetization coincides with one of the
sides of the cube. This direction is called the direction of ready magnet
ization. In order to deflect the magnetic polarization from the direction
of ready magnetization, work must be performed.
A single crystal of a ferromagnetic substance will be magnetized
so that the resultant energy is a minimum equilibrium always
corresponds to minimum energy. However, it does not necessarily
follow from this that all of the single crystal is magnetized in one
direction : in this case it will possess an external magnetic field whose
energy is g \ H 2 dV. This quantity is always positive and increases
the total energy. But if the single crystal is divided into regions
or layers whose magnetization alternates in direction, then the
152 ELECTRODYNAMICS [Part II
external field can be eliminated since neighbouring layers (or, as
they are called, domains) produce fields of opposite sign. In the tran
sition region between domains, the polarization gradually turns from
the direction of ready magnetization in one domain to the reverse
direction in the other domain. Clearly, if a certain direction is the
direction of ready magnetization, then the directly opposite direction
also possesses this property. The structure of the transition region
has been studied theoretically by L. D. Landau and E. M. Lifshits.
The domain structure of crystals was later demonstrated experi
mentally. If a very thin emulsion of particles of a ferromagnetic
substance is spread over the smooth surface of a ferromagnetic single
crystal, the particles will be distributed along lines where the inter
faces between domains intersect the surface of the crystal.
Since, between domains, the polarization is deflected from the
direction of ready magnetization, it is necessary to perform work
to establish the transition region. Summarizing, if the whole single
crystal consists of one domain, its energy increases at the expense
i *
of work done in creating an external field, equal to ^ H z dV\
if, however, the crystal consists of many domains, the energy increases
at the expense of the additional energy of the transition regions.
The equilibrium state will be that state in which the energy is least.
The energy of a field increases with the volume it occupies, that
is, as the cube of the linear dimensions of the crystal. The energy
of the transition regions increases in proportion to their total area.
In a crystal of sufficiently small dimensions, there can exist only
one transition region whose area is proportional to the square of the
linear dimensions of the single crystal. Therefore, in such a small
crystal, the volume energy changes according to a cubic law with
respect to dimensions, while the surface energy varies according
to a square law. In a sufficiently small crystal, the volume energy
becomes less than the surface energy; such a crystal is not separated
into domains but is magnetized as a whole. This has been experi
mentally established in crystals with dimensions of 10~ 4 10~ 6 cm.
The thickness of domains in large crystals of appropriate ferromagnet
ics is of the same order.
An example of the shape of a domain as proposed by Landau
and Lifshits is shown in Fig. 22. The arrows
denote the direction of the polarization in each
domain. The serrations at the boundary almost
completely destroy the external magnetic field ;
Fig. 22 the * nies f magnetic induction inside the crystal
are closed through them, and do not emerge.
The magnetization of a ferromagnetic in an external field. If a
magnetic field is applied to a ferromagnetic crystal in the direction of
ready magnetization, then those regions, for which the polarization
Sec. 16] ELECTRODYNAMICS OF MATERIAL MEDIA 153
is in opposition to the field, are contracted with displacement of
the interfaces and may disappear completely in a comparatively
small field. Then the crystal is magnetized to saturation. In order
to magnetize the crystal to saturation in a direction that is not coin
cident with the initial direction of polarization in the domains, con
siderably larger fields are required.
In a poly crystalline body, such as ordinary steel, the separate
single crystals are oriented more or less at random relative to one
another. In any case, the directions of ready magnetization are not
the same for the separate crystals. When an external magnetic field
is applied, the different crystals are magnetized differently and the
magnetization curve is not as steep as is possible in the case of a
separate crystal. The magnetic interaction between separate crystals
results in a definite magnetic polarization remaining after steel has
been magnetized and the field subsequently removed. This is what
is known as hysteresis.
Magnetic interaction of atoms. Let it be noted, in addition, that
the magnetic interaction between separate elementary (atomic)
magnets is not at all adequate in explaining the cause of ferromagnetism.
The energy of interaction between two elementary magnetic moments
is of the order 10~ 16 erg, while the energy of thermal motion at room
temperature is about 10 14 erg (see Part IV). This is why random
thermal motion should destroy the orderly magnetization already
at a temperature of about 1 above absolute zero. Actually, ferro
magnetism of steel disappears in the neighbourhood of 1,000 above
absolute zero, thus corresponding to an interaction energy between
elementary magnets in the order of 10~ 13 erg.
Ferromagnetism is of quantum origin and cannot be explained
with the aid of classicial analogues.
The relationship between fields and inductions. A substance is
always in equilibrium in a constant external magnetic field. To this
equilibrium there corresponds a very definite induction and polari
zation. In a weak field, the relationship between the quantities is
linear. For this reason, the magnetic induction is expressed linearly
in terms of the magnetic field in a medium:
(16.29a)
In a dielectric, where a static equilibrium polarization corresponds
to a definite electric field, there is a similar relationship for weak
fields.
DsE. (16.29b)
The quantity x is called the permeability and e is the dielectric
constant.
It should be noted that in ferromagnetics the region for which
a linear law is applicable has an upper limit of not very large fields
154 ELECTRODYNAMICS [Part II
(10 2 10 4 CGSE), since saturation sets in; in diamagnetic and
paramagnetic substances at room temperatures a linear law applies
for all actually attainable fields.
The vector nature of electric and magnetic fields. The question
may arise: Why is magnetic induction expressed linearly solely in
terms of magnetic field, while electric induction is expressed solely
in terms of electric field?
In order to answer this question we must examine the vector
properties of electromagnetic quantities in more detail.
Two separate systems of rectangular coordinates exist in space:
a righthand system and a lefthand system. They are related to
each other like left and right hands, if the thumbs are in the direction
of the o;axes, the forefingers aJong the yaxes, and the middle
fingers along the zaxes. It is obvious that no rotation in space can
make these two systems coincide. However, one system transforms
to the other if the signs of the coordinates in it are reversed. Of
course, both coordinate systems are completely equivalent physi
cally. The choice of any one of them is completely arbitrary. Therefore,
the form of any equation expressing a law in electrodynamics should
not change under a transformation from a righthand to a lefthand
system.
Let us now take Maxwell's equation (12.24). In order to perform
a transformation to another coordinate system, it is sufficient to
change the signs of the coordinates. This changes the sign of the
vector operation rot, because this operation denotes a differentiation
with respect to coordinates. What happens, then, to the electric
field components ? Since only one of two vectors is differentiated with
respect to the coordinates, namely E, the sign of one of them must
change in order to retain the form of the equation. It is easy to see
that vector E will change sign. Indeed, the righthand side of equation
div E = 4 TT p is a scalar and does not change in sign. On the left
hand side, the sign of the div operation changes and, hence, the signs
of all the components of E must also change. Therefore, the com
ponents of the magnetic field do not change sign in a transformation
from a lefthand to a righthand system.
In a rotation of a coordinate system, the projections of any vector
are transformed by the same equations of analytical geometry as
the coordinates. As was shown, the change of sign for all three coordi
nates is not equivalent to any rotation. It turns out that some vectors,
such as E, behave quite similarly to a radius vector r; when the signs
of the components of the radius vector r are changed, the signs of
all the components of E also change. Other vectors, such as H, behave
like a radius vector under coordinate rotations, and not like a radius
vector in the transformation from a righthand to a lefthand system.
Vectors that behave like E are called true or polar vectors, while
those behaving like H are called pseudovectors or axia) vectors.
SeC. 16] ELECTRODYNAMICS OF MATERIAL MEDIA 155
Velocity, force, acceleration, current density, and vector potential
are, in addition to the electric field, real vectors while magnetic
moment, angular momentum and angular velocity are pseudo
vectors.
The fact that angular momentum is a pseudo vector can easily be
seen from its definition: M = [rp]. Both factors of the vector prod
uct, r and p, change signs, so that M does not change in sign.
A pseudovector cannot be linearly related to a real vector in electro
dynamics because the sign in any such equality would depend on
the choice of coordinate system, which contradicts physical facts.
For this reason the vectors B and H, D and E appear separately in
the linear laws (16.29a) and (16.29b).
The equations for conductors in a constant field. We shall now
consider the equations of electrodynamics of constant fields for
conductors. As has already been indicated, it is not a constant value
of polarization that is established in a conductor in a constant field,
#P
but a constant rate of increase of polarization =r . This quantity has
O T
the meaning of current density j'. The derivative ^ , appearing on
the righthand side of Maxwell's equations, may be replaced thus:
Tf** 4 J'' < 16  30 >
because the field is constant. Here, the current j' is also continuous.
The magnetic field is a pseudovector and cannot be linearly related
to the current density. We note that for metals the linear relationship
between field and current (Ohm's law) does not break down, no
matter how strong the field.
The quantity cr is called the specific conductance, or conductivity.
In the CGSE system its dimensions are inverse seconds. For metals cr
is of the order 10 17 sec" 1 .
Slowly varying fields. So far, an electromagnetic field in a medium
has been regarded as strictly constant with time. But if the field
varies sufficiently slowly with time, it may also be considered as
constant. Let us give a general criterion whereby we can say what
field may be regarded as slowly varying.
We assume that a constant field is switched on at some initial
instant of time t = 0. A stationary state is not established in the
medium at once but only after a certain interval of time has elapsed.
If, for example, the medium is a dielectric, then, in that time, a def
inite polarization is established corresponding to the given field;
in a metal, characterizes the time taken for a constant current
to be established. is called the relaxation time. If, during the relax
ation time, the field changes by only a small fraction of its value
it can be regarded as constant within the accuracy of that small
fraction. In other words, the criterion of slowness of variation of a
156 ELECTRODYNAMICS [Part II
field is this : within the relaxation time a stationary state, correspond
ing to the given value of field, has time to establish itself in the medium.
Such fields are termed slowly varying. For them, the same values
of permeability, dielectric constant, and conductivity can be sub
stituted into Maxwell's equations, as for constant fields.
Let us write down Maxwell's equations for a slowly varying field in a
conductor. In the expression
c>D 8E #P 3E
the first term can be neglected in the majority of cases because it
in no way exceeds  ; if , as occurs in metals, a is of the order of 10 17 ,
then aE > E/Q. Whence Maxwell's equations are obtained for a
slowly varying field in a conductor:
A. TI **7rj 47TCrJi /ir* fui\
rotH ^ J  = , (16.32)
c c
rotE  A^H (16.33)
C O It
divxH0. (16.34)
This system is complete and, together with the boundary con
ditions (see exercises 1 and 5 in this section), is sufficient for the deter
mination of slowly varying fields in a conductor.
Rapidly varying fields. Let us now consider the case of rapidly
varying fields, i.e., fields which change more rapidly than the relax
ation process or the establishment of a definite stationary stato
in the medium. Then the state of the medium depends not only
on the instantaneous value of the field, but also on its values at
previous instants of time; in other words, it depends on the way
in which the field changes with time. Such a relationship is very
complicated in the general case. It is simplified if the field is weak :
then, at any rate, we may expect the relationship to be linear.
Expansion in harmonic components. Let us examine the general
form of the linear relationship. To do so we represent the field as
follows :
E (t) JTEfeCosfcofc* + 9*) > (16.35)
k
i.e., we expand it in harmonic components. The more values of the
amplitudes E*, frequencies cofc, and phases <pfc, we take, the better
the approximation for the variation of E. However, if the relationship
between induction and field is linear, then the induction is also of
the form of a sum of harmonic components: Dk cos (cok + 9*),
where (and this is most important) each term in the sum for induction
is determined by the term of the same frequency in (16.35).
Soc. 16] ELECTRODYNAMICS OF MATERIAL MEDIA 157
This does not contradict the general statement that the induction
is determined by the entire time dependence of a rapidly varying
field; for a harmonic relationship between field and time, it is fully
given by its amplitude, phase, and frequency. And a component of
the field with a certain frequency can in no way give rise to induction
components with another frequenc3 r if the relationship between
field and induction is linear, for no linear relationship exists between
trigonometric functions of different arguments. Therefore, if we
write the functions even with the same frequency but with different
phases 9^ and fyk, we do not directly obtain a linear relationship
either. However, if we use a complex form and express the field and
induction in terms of exponentials by means of the equations
!()= _
(16.36)
=^a
k
(the star denotes a complex conjugate quantity), then a linear rela
tionship between field and induction can be written in the following
form :
D fe = efeEfc or D (to)  s (to) E (to) . (16.37)
The quantities D& and Efc are complex. From a comparison of
(16.35) and (16.36) it can be seen that they differ from the real field
and induction amplitudes by the complex factors e l *k and   e^k .
Thus, the dielectric constant ek ~ ( to) must also be a complex
quantity. A complex permeability ^ (to) is similarly introduced.
Maxwell's equations in complex form. Let us now write down
Maxwell's equations for complex field components. It must be noted
that they are a function of time according to an e' 63 ' law. We shall
divide the equations by these factors and get the following system
of Maxwell's equations for rapidly varying harmonic fields:
rot II is(co)E, (16.38)
c
rotEi "/(to)H, (16.39)
diveE = 0, (16.40)
div/H 0. (16.41)
The imaginary parts of the dielectric constant and permeability
lead to the energy of a rapidly varying field being spent on the gener
ation of heat in the substance (see exercise 18).
158
ELECTROD Y N AMICS
[Part II
We note, in addition, that for rapidly varying fields the division
of bodies into conductors and dielectrics is conditional and is deter
mined by the relationship between the imaginary and real parts
of the dielectric constant. A substance retains the character of a
conductor up to such frequencies of a rapidly varying field as satisfy
the inequality > 6.
Exercises
1) Show that on tho boundary between two media the tangential compo
nents of tho fields and the normal components of induction are continuous.
Integrate the equations for the inductions over a small flat cylinder,
and tho equations for fields over a narrow quadrilateral bounding the inter
face (Fig. 23a and b).
2) Show that a magnetic field varying sinusoidally with time is damped
with depth in the conductor (x = !)
From equations (16.32) (16.34) we have
dx
d K
~dx
Y _ __ *L
z '
4 77
whence
whore a; is a coordinate, normal to tho surface of a conductor.
3) Show that equations (16.32)( 16.34) aro formally applicable to the
case of rapidly varying fields, if the relationship between field and time is
considered harmonic. In this case the conductivity a is proportional to the
imaginary part of e, and the real part of c is equal to zero.
Fig. 23
b)
4) Calculate the permeability of a substance whose molecules do not possess
intrinsic magnetic moments.
Tho additional velocity of charges, when a magnetic field is applied, is
v [<o r], where <o is given by Larmor's theorem (15.38); whence the magnetic
momont is determined from the general expression (15.19). The mean pro
jection of this moment on H is obtained by averaging over the angles between
H and r. From this wo find the magnetic polarization and, finally, x:
where N is the number of molecules in unit volume and (r l ) 2 ,is the mean
square of the radius of rotation of the t'th charge.
5) Show that an electric field near the surface of a charged conductor is
equal to 4 TC y, whore y is the surface density of static charge on the conductor.
Sec. 16] ELECTRODYNAMICS OF MATERIAL MEDIA 159
We use the same method as in exercise 1 of integrating equation (12.27)
over a small flat cylinder bounding the conductor on both sides, with account
taken of the fact that the field inside the conductor is equal to zero. For quasi
stationary fields ^ = /', where j' n is the projection of ]' on the external
ot
normal.
6) Calculate the energy of a system of charged conductors in a vacuum.
Tfl2
We substitute E = v <P in the definition of energy & eleotr = I d V
and integrate by parts, taking advantage of the fact that in a vacuum
A 9 = 0. Since, from exorcise 5 in this section, the field close to a conductor
is equal to 4 TT y , and the surface of a conductor is equipotential, wo reduce
to the form
& electr = ~tT/, e * W
i
where e\ is the charge on the tth conductor arid 9; is its potential.
7) Determine how a constant uniform electric field changes if a conducting
sphere is introduced into it.
The field potential must be sought in the form 9 = li! r 1 , where
the vector d is in tho same direction as the initial uniform field E . d is deter
mined from the condition that the tangential component E y 9 on
the sphere is equal to zero.
8) Determine in what way a constant uniform electric field E varies if
a dielectric sphere of dielectric constant e is introduced into it.
The field potential outside the sphere must bo sought in the same form
as in problem 7, but, inside the sphere, it must be sought in tho form E' r,
where E' const. Determine tho vectors d and E' from tho boundary condi
tions derived in exerciso 1.
9) Find the electric field which arises in space when a point charge e is
brought to a distance a from an infinite flat conducting surface.
We drop a perpendicular from the point at which the charge is situated
to the surface of tho conductor and, at a distance a inside the conductor, wo
place an equal and opposite (fictitious) charge e. Then, the field component
tangential to the surface becomes zero. Tho field outside the conductor is
everywhere equal to the vector sum of tho fields due to the real and fictitious
charges.
10) Find the electric field when a point charge e is brought to a distance
a from an infinite flat surface of a dielectric with constant e. The dielectric
is infinitely deep.
We make the same construction as in exercise 9. We look for the field
e' r
inside the dielectric of the form ~, where FJ is a radius vector from tho point
r i
at which the real charge is situated; the field outside the dielectric is of the
c TI &" r
form 3 + o where r 2 is a radius vector drawn from the "image" of the
r \ ^a
charge. The constants e' and e," are determined from the boundary conditions
of exercise 1.
11) Assuming that x 1> determine the magnetic energy of a system
of conductors carrying a steady current.
Starting from the equation $ mag = I H 2 d V t we substitute H = rot A
O7T J
and integrate by parts using (11.29). The surface integral at infinity is equal
160 ELECTRODYNAMICS [Part II
to zero. Then we use (16.32) and reduce ^mag to the form
12) Express the magnetic energy for a system of currents in terms of a
double integral over the volumes of the conductors.
We replace the summation over the charges, in equation (15.13), by a
volume integration. This gives
1 ff IWJ(r')
= ! c f JJ T^:^
For line conductors wo can substitute / di instead of \dV, provided dV
and dV' are volume elements of different conductors. Then the mutual magnetic
energy for two line conductors i, k is
& tk mns '  ' o
whoro  r, rk \ is the distance between the elements of contours d\i and
d\k. When i = k we must regard the conductor as thin, though not infinitely
thin, otherwise the integral diverges. The intrinsic magnetic energy for one
conductor is
j\/tft is called the mutual induction coefficient, and MU is the selfinduction
coefficient.
13) Writo down the Lagraiigian for a system of currents, assuming that
thoro is capacity coupling between the conductors by virtue of capacitors
conrioctocl in the circuit.
Because of the linearity of oloctrodynamical equations, the potential of
the ith conductor is expressed linearly in terms of the charges on all the con
ductors:
92;
k
From exorcise 6, we obtain the electric energy.
_ 1 VV
G elect r ~ ~7T / , ^ik &i &k
Tho charge on the plates of a capacitor is related to the incoming current
by 6fc  /fc. With the aid of exorcise 12 we obtain the magnetic energy
I V^T 1 \^r
i.fc " i,k
From See. 13 the Lagrangian (neglecting sign) is (H  E 2 ) dV, whence
O 7C J
L  o mafif o electr ~j~ / ,
[cf. (17.16)].
Sec. 16] ELECTRODYNAMICS OF MATERIAL MEDIA 161
14) Determine the work performed in unit time by a varying electromagnetic
field in a medium.
We write equation (13.23) for the external space not occupied by the sub
stance, where p = 0. From the boundary conditions of exercise 1, it follows
that the normal component of the Poynting vector U is continuous at the
boundary of the body. From this, applying the same transformations to equa
tions (16.17) and (16.25), as load to (13.23), we find
where = may be expressed in terms of the change in energy of the field in
d t
the external space.
15) Calculate the energy transformed into heat in unit time in a conductor
situated in a constant field. Assume H to be a singlevalued function of B.
o o
For such a body, where H and B are related uniquely, H jr = ^ / (B) ,
B*
where / (B) is some function of B. For example, for B = y. H, / =*  > . The
result of the preceding t3xercise gives
"* f/(BL dF _ 1
tit dt J 4 TT 4 TT
see (16.30) and (16.31). For a constant field, there is zero on the lefthand
side of the equation, while the righthand side is an essentially positive quantity.
This energy must therefore bo converted into heat according to the energy
conservation law.
16) AVrito down Lagrange's equation for a system of currents taking into
account the conversion of energy into heat.
From exercise 14, the hoat generated in unit time may be written as
y^rj Ij, where ri is the resistance of the i"th conductor. Wo search for Lag
range's equation with the righthand side in the form
d d L d L
7 ^~. r = v; .
at oe{ <J&i
From the definition (4.4) we find  as
Ci t
d dL
. ^ > t>: I
dt
Whence v, = n Ii ~ r; ei .
17) Reasoning in the same way as for exercise 15, show that if H is a double
O Trt O n
valued function of B, having one value for ^ < and the other for > 0,
d t c> t
then, for a periodic variation of B, tho heat generated in one period is equal
TT /I tt
r , where the integral is taken over one period.
to f
18) Show that if e (w) and x () possess imaginary parts, heat is generated
in a rapidly varying field.
11 0060
162 ELECTRODYNAMICS [Part IT
Tho density of heat generation may bo represented as the divergence from
^
the Poynting vector U ==: [E H] . In forming quadratic complex quan
tities, we must tako into account their time dependence. For example, if we
take E and H with a factor e~w, their product will be proportional to e~2 >'.
After time averaging, this factor will yield zero. Therefore wo must take only
products of the form [EH*J 4 [E*H]. Now, u.sing equations (16.38) and
(16.39), wo obtain
~ cliv ([E* II] 4 [E H*]) =  / 1 6i (c  e*) EE* 4 ~ ico (/  /*) HH* .
4 7C TC TC *x <v
Hero, both parts of the equation are real, and if e ~ GJ 4 ify an( l X ^ 7.\ + * X
then on tho righthand side thore is the expression o (e 2 EE* h X2 HH*) .
From this it can bo soon that e 2 > and x 2 > 0, since tho energy of tho field
is absorbed by tho modiutn.
19) Calculate tho dielectric constant of a medium, considering that all
tho charges in it are connected by elastic forces with tho equilibrium positions.
The characteristic oscillation frequency of tho charges is arid tho frequency
of tho field is o>.
Tho radius vector of a charge satisfies the differential equation
m ( i 4 r)  e E e i = e E .
Its solution has tho frequency of tho external field and may bo written as
eE
r =
0> 2 )
Tho polarization can bo obtained from this by multiplying by tho number
of charges in unit volume N and by e. Since tho induction D is equal to e E
4 TT N e 2
or E 4 4 TT I*, wo find that e = 1 4 , = . For w * 0, wo obtain the
m(wj w)
static dielectric constant e 1 4 For co > oo, e (to) is obtained
?no> \ /
r r r i 4 * N e *
for very largo frequencies or for Iroo charges e = 1  2  .
Sec 17. Plane Electromagnetic Waves
General equations. In this section we shall first consider the solutions
of Maxwell's equations for free space, i.e., in the absence of charges.
These solutions, as we shall see, are of the form of travelling waves.
Analogous solutions also exist for a nonabsorbing material medium.
These solutions will also be found in the present section.
In the absence of charges or currents, the equations for scalar
and vector potentials are written thus:
0, (17.1)
A9^ = 0, (17.2)
with the additional condition (12.36)
Sec. 17] PLANE ELECTROMAGNETIC WAVES 163
divA + 4^0. (17.3)
Equations of the form (17.1) or (17.2) are called wave equations.
The solution of a wave equation. We shall look for particular solu
tions of equations (17.1) and (17.2) which depend only 011 one coordi
nate (for example, x) and on time. Then the wave equations can be
rewritten in the following manner:
and the supplementary condition takes the form
^ + _L. =0 . (17.6)
dx c dt v '
We shall now find the solution of (17.4) or (17.5) without imposing
any further restrictions. We shall temporarily introduce the following
notation :
x + ct = %,
(17.7)
X Cl = 7) .
We transform (17.5) to these independent variables. (17.5) can be
rewritten symbolically as
Then
9 _ c>9 d #9 dt\ #9 #9
~$x = T IT ~d#" + ~^r "^ = ^T ~^T *
JL_^. ^L Ji^l _i_ _^JL^ZL _?. __ 5<p
T~aT "" "aT"c""a? + "^T"c""a7 ~" "THf "" "a^"'
because, for constant ^ (i.e., d^ = 0), ^ = 1 and T~ = 1, while
<7 37 G X
for constant # (i.e., dx = 0), ^ =  rrf = 1 in accordance with
c ot c ot
the same equations. Thus, symbolically
_i_ + J__L ==2 A A__LA = 9_i_.
^a? ^ c dt b^^ 'dx c r ot ~ c^ '
i a w a__ i g \ __ & 9 _
a? ^ c dtl\dx c dtjV ~ d^dri ~~ '
Hence, wave equations (17.4) and (17.5) are written thus:
164 ELECTRODYNAMICS [Part II
Integrating any of them "with respect to , we obtain
~ = C (',), %C'W (17.10)
It is not difiicult now to integrate with respect to TJ:
f, r i
A  J C (/)) d v) + C\ (5) , 9 ^ / C" (/)) cZvj + t; (5) .
Finally, the required solution is written as:
AA^YjHA^), 99i(*iH9a(5), (17.11)
since the substitution of (17.11) into (17.9) gives zero identically.
Passing to the variables it*, /, we can write the solutions to (17.4),
(17.5):
A = A! (a; c) + A 2 ( x + r O > 9 = 9i (* C + 9 2 (a + cl) . (17.12)
Plane travelling waves. The solution depending on x+ct does
not depend on the solution whose argument is x c I ; these are
two linearly independent solutions. Therefore it is sufficient to con
sider one of them :
A = A(*cO, (17.13)
9 = 9(.r <jf). (17.14)
In order to satisfy the supplementary condition (17.6), we perform
a gauge transformation:
9 (.i;.fO = 9 f ^cOy~./(a:cO = 9' + / (17.15)
(the dot over / denotes differentiation with respect to the whole
argument Y) = x c t). But, if we put 9' = /, we obtain simply
9^0. Then, from (17.6), we also obtain A x = 0. Thus, for a solution
of the form considered, depending on x ct only, the Lorenz con
dition is satisfied most simply by substituting 9=0, A x 0.
The electric field component x is equal to zero:
From the general result of Sec. 12, this property of E x does not
depend on a potential gauge transformation.
The magnetic field component x is also equal to zero:
Sec. 17]
PLANE ELECTROMAGNETIC WAVES
165
We find the remaining field components:
F _ 1 dA Y __ : ^ __ 1 &A __ ;
Ay ~"""7~ar~^ yj A;r ~"T~^T"^^ 3
(17.18)
From this equation it follows that E and H are perpendicular,
because
EH =fiy#y + *#*=<>. (17.19)
They are equal in absolute magnitude, since E = H \/A* + A%.
The solution of the form (17.13) has a simple physical meaning.
Let us take the value of E at an instant of time t =~ on the plane
x = 0. It is equal to E (0). Tt is clear that the E (0) will have the
samo value at the instant of time t on the plane x = c t, because
E (x ct) E (0) on that plane. We can also say that the plane
on which the field E is equal to E (0) is translated in space through
a distance ct in a time /, i.e., it moves with a velocity c. The samo
applies to any plane x = X Q , for which there was some value of field
E (x ) at the initial instant of time. To summarize, all planes with
the given value of iield are propagated in space with velocity c.
Therefore, the solution E (# ct) is called a travelling plane wave.
We note that the form of the wave docs not change as it moves;
the distance between planes x = x l and x = # 2 , for which K is equal
to E (XL) and E (# 2 ), is constant. This result
holds for any arbitrary form of wave, pro
vided it is travelling in free space.
Repeating, the velocity of propagation
of a wave in empty space does not depend
on its shape or amplitude and it is equal to
a universal constant c.
The transverse nature of waves. The elec
tric and magnetic fields, as we have seen
from (17.19), are perpendicular to the
direction of wave propagation, as well as to
each other. This is why it is said that elec
tromagnetic waves are transverse (as op
posed to longitudinal sound waves in air, for
which the oscillations occur in the direction of propagation). The
direction of propagation, the electric field, and the magnetic field are
shown in Fig. 24. In it, n is a unit vector along the #axis.
In future it will be sufficient to take only one component of the
electric field. For this it is necessary to take one of the coordinate
axes, for example the ?/axis, in the direction of the electric field,
which in no way limits the generality. This is shown in Fig. 24.
Fig. 24
166 ELECTRODYNAMICS [Part II
The ^coordinate will be written in the form x rn, so that
E y = A y (rn ct) . (17.20)
But in this notation it is not necessary to relate the vector n, in the
direction of propagation to the #axis. A solution with argument
of the form (17.20) is applicable to any direction of n, provided,
naturally, that n, E, and H are mutually perpendicular.
The momentum density of the wave [see (13.27)] is equal to
and is directed along n. The energy density is
87C ~ 47T
It differs from the momentum density by the factor c. This, as
we shall see later, is very essential for the quantum theory of
light.
Pressure ol light. If a wave falls on an absorbing obstacle, for
example, 011 a black wall, and is not reflected, then its momentum
is transmitted to the wall in accordance with the conservation law.
But momentum transmitted to a body in unit time is, by Newton's
Second Law, nothing other than force. It follows that there is a force
A 2
of  7 for every square centimetre of the absorbing barrier, upon
which the wave is normally incident. Force referred to unit surface
is, by definition, the pressure of the electromagnetic wave on the
barrier. Consequently, electrodynamics predicts the existence of
light pressure. This was observed and measured by P. N. Lebedev.
Harmonic waves. A special interest is attached to travelling waves
for which the function E (x ct) is harmonic. The most general
harmonic solution is of the following form:
B=Ro{Pc"'"( < " I r)} > (17.21)
where the symbol Re {} denotes the real part of the expression
inside the braces, F is a complex vector of the form Fj + i F 2 [Cf.
(7.14c)], and co is the wave frequency in the same sense as in equation
(7.3). co is the number of radians per second by which the argument
of the exponential function changes.
The wave vector. The vector co is called the wave vector. It is
c
denoted by the letter k:
k^to^. (17.22)
SeC. 17] PLANE ELECTROMAGNETIC WAVES 107
The geometric meaning of k is easy to explain. We define the wave
length, i.e., the distance & rn in space at which E assumes the same
value. Let the required wavelength be X. Then
A Arn
I (0  JO)  27TI ,,_ Xrt
e < = e c =e , (17.23)
because the period of the function e ix is equal to 2 n. Hence,
X= 2  . (17.24)
CO
Comparing the wavelength with the wave vector, we obtain
k = ^n, X^ . (17.25)
A 1C
Polarization of a plane harmonic wave. Let us now study the
nature of the oscillations of an electric field. To do this, we write the
vector F in the form
F = l\ + il\ : (! i Eo) e'' a . (17.26)
We choose the phase a so that the vectors E t and E 2 are mutually
perpendicular. We multiply equation (17.20) by e~ i<x and square.
Then we obtain
(E x ;E 2 ) 2 = ElE** = e 2l *(Fl  Fl + 2i(P 1 F 2 )) . (17.27)
We have taken advantage of the fact that Ej and E 2 are perpendic
ular. Because of this (E x fcE 2 ) 2 is a purely real quantity. Therefore,
the imaginary part of the righthand side of expression (17.27) must
be piit equal to zero. Representing e~ 2lot as cos 2 a i sin 2 a, we
obtain
 (Fl  Fl) sin 2a + 2 (F! F 2 ) cos 2 a ==
or
whence the angle a is determined for the given solution (17.21).
It is now easy to express E x and E 2 . Indeed, from (17.26), EJL
E 2 = (F t + iY 2 ) e a = Fj cos a + F 2 sin a i (t\ sin a F 2 cos a),
so that
E, = FT cos a + Fo sin a , 1
[ (17.29)
E 2 = F x sin a F 2 cos a . J
We now include a constant phase in the exponent (17.21) and, for
short, put
a _ co j _ 2JLJ == <, . (17.30)
163 ELECTRODYNAMICS [Part TI
Then, in the most general case, the electric field for a plane harmonic
wave will be
E = Re {(E x  i E 2 ) e'*} = E 1 cos fy + E 2 sin ty . (17.31)
Here, the vectors E t and E 2 are defined as perpendicular.
Let us assume that a wave is propagated along the .zaxis.
The ?/axis is directed along E 1? and the zaxis along E 2 . Hence, from
(17.31), we obtain
E Y = E 1 cos fy , A 1 * = # 2 sin ^ . (17.32)
Let us eliminate the phase <. We divide the first equation by K l ,
the second by E 2 , square and add. Then the phase is eliminated and
an equation relating the field components remains:
* 4 ~ 1 /i 7 wi
K\ + El ~ L ' lU^J
It follows that tho electric field vector describes an ellipse in the
//zplane moving along tho #axis with velocity c, and passes round
the whole ellipse on one wavelength. Relative to a fixed coordinate
system, the electric field vector describes a helix wound on an elliptic
cylinder. The pitch of the helix is equal to the wavelength.
Such an electromagnetic wave is termed ellipticallv polarized. It
represents the most general form of a plane harmonic wave (17.21).
If one of the components is equal to zero, for example Ej or
E 2 = 0, then the oscillations of E occur in one plane. Such a wave
is termed plane polarized.
When EI is equal to E 2 , the vector E describes a circle in the
?/;:plane. Depending on the sign of E z , the rotation around the
circle occurs in a clockwise or anti
clockwise direction. Accordingly,
the wave is termed righthanded
or lefthanded polarized. These
waves are shown in Fig. 25. For
the same value of phase i, the
rotation occurs either in a clockwise
or anticlockwise direction.
The sum of two waves of equal
amplitude, which are circularly
polarized, gives a plane polarized wave. The relationship between
their phases determines the plane of polarization. Thus, if the waves
shown in Fig. 25 are added, the oscillations E 2 and E 2 mutually
cancel and only the plane polarized oscillation E x remains.
In turn, a circularly polarized oscillation is resolved into two
mutually perpendicular plane oscillations.
Certain crystals, for example tourmaline, are capable of polarizing
light.
SeC. 17] PLANE ELECTROMAGNETIC WAVES 109
Unpolarized light. In nature, it is most common to observe un
polarized (natural) light. Naturally, such light cannot be strictly
monochromatic (i.e., possessing strictly one frequency co), for, as we
have just shown, monochromatic light is always polarized in some
way. But if we imagine that the components E t and E 2 in Fig. 25
are not related by a strict phase relationship (17.32), but randomly
change their relative phases, then the resultant vector will also
change its direction in a random manner. However, for this, it is
necessary that the oscillation frequencies should vary in time within
some interval A co, since the difference of phase between two oscil
lations of strictly constant and identical frequency is constant.
The propagation of light in a medium. We shall now consider the
question of the propagation of light in a material medium. At the
end of the preceding section we said that the quantities z and ^
have meaning only for oscillations of a definite frequency <o. To
simplify notation, we shall not use the symbol for a real part Re {},
remembering that the real part is always taken. Since all the quanti
ties depend on time according to an c >6) ' law, the derivative I
() t
reduces to a multiplication by iv. Then the system of Maxwell's
equations can be written in the following form:
rotH= l?LeE, (17.34)
(17.35)
0. (17.37)
Once again we look for a solution in the form of a plane wave.
Since the time relationship is already eliminated, all the quantities
depend only on one coordinate, for example, upon x. From (17.35)
and (17.37), it follows that
or
because a solution that is constant over all space does not represent
any wave. Thus, the waves are transverse. Equations (17.34) to
(17.37) are satisfied if we substitute E y = E (x), E Z = Q, H y = 0,
H Z H (x), or, in other words, if the electric field is directed along
the yaxis and the magnetic field along the zaxis (a righthanded
system).
170 ELECTRODYNAMICS [Part II
Indeed, there then remain the following equations:
**L=*2sE, (17.38)
(17.39)
Eliminating any of the quantities E or H, we obtain equations
which are identical in form. For example
whence
E = Fe*T v ** *"*'. (17.41)
If the wave is propagated in any arbitrary direction, and not
along the #axis, then the solution (17.41) is rewritten thus [here,
the symbol Re {} is included for a comparison with (17.21)]:
# = RepV~' w V r vex ;j. (17.42)
And so, compared with (17.21), the wave velocity has been multi
plied by ~T / . Accordingly, the wave vector will, instead of (17.22),
formally satisfy the equation
k = Vsx W (17.43)
However,  '  is the velocity of light in a medium, and (17.43)
is the wave vector, provided e and % are real numbers. Then equation
(17.43) will be fully analogous to (17.25). In this case, the solution
(17.42) is periodic in space and in time and describes a plane wave
travelling with velocity ^=
vx
The ratio of the wave velocity in free space to that in a medium is
called the refractive index of the medium for waves of given frequency
co. We note that for visiblelight frequencies, e (co) has nothing in
common with its static value. For example, water has a dielectric
constant of 81, so that v's = N/81 =9, while the refractive index in
the visible frequencies is approximately equal to 1.33 (x. can be consid
ered equal to 1).
Absorption of light in a medium. We now consider a more general
case of complex e = S! + s a (for simplicity we shall put X= 1). As was
shown in exercise 18, Sec. 16, the imaginary part of s accounts for
absorption of light. We shall denote the root of the complex dielectric
constant thus :
Sec. 17]
PLANE ELECTBOMAGNETIC WAVES
171
VT = Vsj + i e 2 = v x + z v 2 . (17.44)
Let us substitute this expression into the exponent of equation
(17.42), putting 7i x ~l, n y = 0, n z 0. Since t 2 1, wo obtain
/ .XVj \ .VVgtO
E = Ko{Fe"" l< ^ < '{e"" 7 "". (17.45)
Thus, the wave is damped in propagation.
Its amplitude diminishes e times at a distance . A solution of the
cov 2
form (17.45) cannot exist in a region which extends to infinity in all
directions because x= oo substituted into (17.45) yields E oo.
A solution which is damped in space can be used, for example, when
an electromagnetic wave from free space is incident on an absorbing
medium. And the ;raxis in (17.45) must be considered as directed into
the medium.
Exercises
1) Consider tho reflection of a piano electromagnetic wavo from tho inter
face between two transparent (nonabsorbing) media a and 6 with refractive
indexes v ld s= v rf and v^ ^ v& (v 2d = v 2 & = 0). Solve the problem in two
cases: 1) the electric vector lies in the plane drawn through tho normal to the
interface and through tho wavo vector k, II) tho electric vector is parallel
Fig. 20
to the interface between the two media. Calling tho angle of incidence and
the angle of refraction &, find the ratio of tho amplitudes of tho incident and
reflected waves for both cases, I and II (Fig. 26).
At the interface, the normal components of the inductions and the tangential
components of tho fields must bo equal. In order to satisfy the conditions at
the boundary, it is necessary to introduce a third wave, which is reflected
from the interface. We shall take tho equation of tho interface to be y = 0.
The phase of the incident wave at the interface is 2 n x x = ~ x sin 8, that
c c
for the reflected wave x sin 6 X , and that for tho refracted wavo x sin l> .
All three phases must coincide over the whole interface, whence
v rt sin = v& sin & (the law of refraction),
= 0! (the law of reflection).
172 ELECTRODYNAMICS [Part II
Taking into account that // = v E [this is easily obtained from (17.39)
and (17.41)], wo write iho boundary conditions (see exorcise 1, Sec. 1*5):
V* (E sin #! sin 6) = vg E 2 sin $ ,
E cos  E l cos E z cos 0 ,
where E, A\ and / 2 are the electric fields in the incident, reflected and retract
ed wavoH. We can see that, by eliminating E, l^ l and E from those conditions,
\vo again obtain Iho law of refraction. The ratio of tho amplitudes is
E
tanJOj J))^ j^
larT(oT"#j * * '
If f ft = ~ , then A\ and reflection does not occur. (How can this
be verified by double reflection ?)
Tn case If wo must write down tho boundary conditions and obtain the
equation
'" ,_ .).. (ID
sin (0 { ft) v '
(1) and (ft) are called tho Fresno! equations.
2) In tho case  sin 0> 1 (total internal reflection), show that instead
of equation (L) and (IT), a roHoctinn coofficient of unity is obtained. Find
at what depth tho wave, passing in the medium &, is attenuated e times.
3) Kind tho frequency of electromagnetic oscillations in an infinite square
prism with perfectly reflecting walls, assuming a longitudinal electric field
constant along the length of tho prism. Consider that tho field inside tho prism
does not become zero.
Wo must consider that tho tangential component of the electric Hold at
the walls of tho prism is equal to zero, so that tho normal component of tho
I'oynting \ector U should become zero. Tho solution to Maxwell's equations
can 1)0 obtained from tho potential A x A sin sin^e iw , A y ,4~ =
^rp ^= (tho .reoordinato is taken along the axis of the prism). It is of the form
E x E sin sin z e~ <wf ,
Jf r T "*% TC If iwf
H v = II Q cos sin e ,
1 a a
cos 
a
on tho condition that
E = //o and o> 3 = ^ c a
(a is tho side of the square).
4) Solvo the same problem for a travelling wave in the prism (a waveguide).
Tho field in a waveguide is not zero anywhere except at the walls.
The form of tho vector potential in the previous problem suggests one of
the following possible solutions:
Sec. 18] TRANSMISSION OF SIGNALS. ALMOST PLANE WAVES 173
A x = A^e '' "**> sin ^ sin ,
a a
A y = A w e 1 <**> cos 2L sin ,
d Ql
A z = 9 = .
In order to satisfy the condition div A = 0, we must demand that
ikA ox A oy 0.
The normal (to the walls) components of the vector U are again equal to zero
since E X = Q at the walls.
We determine the frequency from equation (12.37):
From here it can bo seen that the frequency obtained in. the .example 3) is
the smallest wave frequency that can be propagated in the prism. This wave
corresponds to X = oo.
5) Show that when two waves, which are circularly polarized in opposite
directions, and which have equal amplitudes (but with frequencies differing
by a small quantity Ao>) and are travelling in one direction, are combined,
a wave is obtained whoso polarization vector rotates to an extent depending
on the distance of propagation.
Sec. 18. Transmission of Signals.
Almost Plane Waves
The impossibility of transmitting a signal by means of a mono
chromatic wave. A plane monochromatic wave (17.42) extends with
out limit in all directions of space and in time. Nowhere, so to speak,
does it have a beginning or an end. What is more, its properties are
everywhere always the same ; its frequency, amplitude and the distance
between two travelling crests (i.e., the wavelength X) are always
constant. All this can be easily seen by considering a sinusoid or helix.
Let us now pose the problem of the possibility of transmitting an
electromagnetic signal over a distance. In order to transmit the signal,
an electromagnetic disturbance must be concentrated in a certain
volume. By propagation, this disturbance can reach another region of
space: detected by some means (for example, a radio receiver), it will
transmit to the point of reception a signal about an event occurring
at the point of transmission. Likewise, our visual perceptions are a
continuous recording of electromagnetic (light) disturbances origi
nating in surrounding objects.
A signal must somehow be bounded in time in order to give notice
of the beginning and end of any event.
In order to transmit a signal the amplitude of the wave must, for
a time, be somehow changed. For example, the amplitude of one of the
waves of a sinusoid must be increased and we must wait until this
increased amplitude arrives at the receiving device. A strictly mono
chromatic wave, i.e., a sinusoid, has the same amplitude everywhere
174 ELECTKOD YNAMICS [Part II
and is therefore not suitable for the transmission of signals in time.
In the same way, an ideal plane wave with a given wave vector cannot
transmit the image of an object limited in space.
The propagation of a nonmonochromatic wave. We shall now con
sider what can be done by superimposing several sinusoids upon one
another. Suppose that the frequencies of all these travelling waves
are included within an interval co ^ co <: co \ 5 . We shall
consider that the frequency interval A co is considerably smaller than
the "carrier" frequency co . The amplitudes of all the waves E Q (co)
will be assumed to be identical for any frequency within the chosen
interval, and equal to zero outside that interval.
Then the resultant oscillation will be represented by the integral
of all the partial oscillations :
to + A to/2
E = f E Q (co) e~ i( ^~ kx) d co = EQ j eT* (6i '* v) rfco . (18.1)
6) A 6)/2
In this equation not only the frequency is variable, but also the
absolute value of the wave vector k (the socalled wave number).
According to (17.22), it is equal to in free space and, in a material
c
medium, k =^%, where v in turn is a function of frequency. In future,
in this section, we shall always assume v 2 0, i.e., that there is no
absorption.
Since the frequency lies within a small interval, k can be expanded
in a series in powers of co co :
k (co) = k (co ) + (co co ) ~~J . (18.2)
Substituting (18.2) in the integral (18.1), we obtain the following
expression for the resultant field:
Aco
Aco
<o u ^ 
We now introduce a new integration variable = co co . Then the
integration can be easily performed and the field reduces to the follow
ing form:
A co/2
A co/2
. if ldk
* 7
\ I Aco i
I .r  
kJ^ JL . (18.4)
x
Sec. 18]
TRANSMISSION OF SIGNALS. ALMOST PLANE WAVES
175
The shape of the signal. Let us now examine the expression obtained.
It consists of two factors. The first of them, J0 e (<0 ' **>, repre
sents a travelling wave homogeneous in space with a mean * 'carrier"
frequency w . However, the amplitude of the resultant wave is no
longer constant in space because of the second factor:
2l 6111 \ I t ~~~ I j I it I 
U U"/Q J
, l dk \
t  1j I
__ I 1 1 \
f[. ldk\
s=s
where the designation g and fy are obvious from the equations. This
factor has a greatest maximum at a;=(^J t y i.e., when the argument
of the sine and the denominator are equal to zero. The other extremes
will be the less the greater their number
(Fig. 27). The greatest maximum is equal
to A co (since 5^1 =1 f or i; = 0). This
maximum is not situated at a fixed place,
but moves in space with a velocity
ffW
V = yy
dk
(18.5)
because, from the definition of the point
of maximum = Q, it follows that
~~JT
dk
Tt
Fig. 27
As we indicated at the beginning of this section, a signal can be
transmitted from one point of space to another by means of a displace
ment of the maximum, since this maximum is distinguished from other
maxima.
A disturbance of this kind concentrated in space is called a wave,
packet.
The propagation of a signal of arbitrary shape. A wave packet need
not necessarily have the form shown in Fig. 27. By choosing a rela
tionship for E Q (to) other than that in equation (18.1) (i.e., by choosing
not a constant amplitude in the interval Aw, but a more complicated
frequency function), the shape of g (^) can be changed. For instance,
the resultant amplitude may have the shape of a rectangle, so that
the transmitted signal will resemble the dash in the Morse code. If
the frequency w is within the radio frequency range, then the signals
can follow upon one another within audio frequency, in this way repro
ducing music or speech.
The frequency range and the duration of the signal. In order to trans
mit a signal, it is always necessary to choose a range of frequencies.
176 KLECTKODYNAMICS [Part It
Let us determine this range. Suppose that the receiving device is
situated at some point x = const. The width of the received signal
can be seen from Fig. 27. In units of <>, it is equal to n in order of magni
tude. Therefore, the duration of the signal is determined from the
equation
In other words, the duration of the signal A is related to the frequency
interval Aco necessary for its transmission by the expression
AcoA~ 2n. (18.6)
It should be noted that this estimate refers only to the order of
magnitude of A to and A. The determination of A <p is, to some extent,
arbitrary. In certain cases AeoA > 2 TU, so that the estimate (18.6)
is a lower figure.
If a radio station is required to transmit sounds audible to the human
ear, then the quantity A is not greater than 0.5 x 10~ 4 sec, since
the limit of audibility is 2 x 10 4 oscillations per sec. From this
The range of Aco is always less than the "carrier" frequency co
which, even for the longestwave transmitting stations, is not less
than 1.5 x 10 5 x 2 TT. In practice, an interval of A to three or four times
less than the value given is quite sufficient, since clipping off the
very highest frequencies in music, singing or speech does not introduce
any essential distortion.
Television transmissions require a considerably greater frequency
interval, because an image must be reproduced 25 times every second;
and, in turn, the image consists of tens of thousands of separate signals
(points). As a result, the carrier frequency is about 2?cx6x 10 7 ,
corresponding to the metric band of radio waves. Such waves are prop
agated over a relatively small radius. They are screened by the cur
vature of the earth's surface like light. The relation (18.6) is always
correct in order of magnitude; therefore, for distant television trans
missions, it is necessary to have either relay stations, very highplaced
transmitters, or cable lines.
Phase and group velocity. We shall now consider in more detail the
velocity with which signals are transmitted. From (18.5), the velocity
of a wave packet is
It differs from the propagation velocity of the constant phase sur
face, which is expressed in terms of frequency and wave number as
Sec. 18] TRANSMISSION OF SIGNALS. ALMOST PLANE WAVES 177
Indeed, the expression for a travelling monochromatic wave can be
written in the following form :
Comparing this formula with the general expression for a travelling
wave E = E (x ut), we arrive at (18.7). The velocity of the wave is u,
and not c, because (18.7) is by no means necessarily related to the pro
pagation of a plane wave in a vacuum.
u = ~ is called the phase velocity of the wave ; v is called the group
velocity of the wave packet obtained by superimposing a group of
waves. In a vacuum, v and u coincide because to ck. However, if
there is dispersion, i.e., a dependence of the refractive index on the
frequency, then co = k so that v^k.
The group velocity may be regarded as the velocity of propagation
of a signal only when it is less than the velocity of light in free space c.
If the expression (18.5) formally gives v > c, we cannot avoid a more
careful analysis that takes into account absorption. As a result, it
turns out that an electromagnetic signal in the form of a very weak
precursor is propagated with a velocity c, but the major portion of the
wave energy arrives at the point of reception with a lesser velocity
(see A. Sommerfeld, Optik, Wiesbaden, 1950).
As an example of the calculation of group velocity we shall take the
dependence of frequency on the wave vector in the form :
to 2 = a 2 + c 2 & 2 .
This form is obtained for a waveguide (see exercise 4, Sec. 17, or
exercise 19, Sec. 16, in the limiting case of extremely large frequencies).
Whence the group velocity is
CO
and since ck < to, we have v < c.
Here, the phase velocity proves to be greater than c:
CO CO
U= j = C 7 >C .
k ck
We note that uv='c*.
In vector form the group velocity is defined as follows :
T. (188)
If we make use of a more accurate dispersion law (obtained in exer
cise 19, Sec. 16), then for to 2 ~to 2 there proves to be a frequency
region for which e (co) is negative. For such frequencies the refractive
index is purely imaginary and the expressions (18.5) and (18.8) become
meaningless.
120060
178 ELECTRODYNAMICS [Part II
The form ol a wave in space and the range of the wave vectors. An
expression similar to (18.6) can also be obtained for the form of a wave
in space at a definite instant of time. For this we must put t = const
and then, once again taking A<p ^TU, we obtain
. , AG> dk
Afc.Az~ 27u. (18.9)
This means that if we want to limit the extent of an electromagnetic
disturbance to a region A#, we must perform a superposition of mono
chromatic waves in the interval of values k of order r^ . In three
Arc
dimensions (18.9) is rewritten thus:
&k x  A#~ 2n ,
AiyA/~ 2n, (18.10)
The limiting accuracy of radiolocation. We shall explain the relations
(18.10) by means of a graphic example. Let us suppose that an electro
magnetic wave has, in some way, to be bounded on the sides, as in the
case of a radiolocation (radar) beam. Let us find the greatest accuracy
with which the locator can register the position of an object at a dis
tance I. Obviously, this accuracy is given by the transverse diameter
of the beam d at a distance I from the locator.
Let the frequency at which the locator works be equal to co, then
the corresponding wavelength is X = ^ . If the electromagnetic wave
were to be propagated in unbounded space it would have an accu
rately defined wave vector
(n is a unit vector in the direction of the beam). If the wave has a
cross section d, then k can 110 longer be regarded as an accurately
defined vector along n. In order to write down an expression for the
electromagnetic wave at any point in space occupied by the beam,
it is necessary to take a group of plane waves whose vectors k lie inside
a cone described by a certain angle of flare. The maximum deviation
of the wave vectors of these plane waves from the mean vector k,
determined from (18.11), will be called k . Here, we do not have in
mind a cone with a sharply bounded surface, but only an estimate
of the angular flare of the beam. According to (18.10) k is related to the
whole cross section of the beam by the following relation :
7u. (18.12)
Sec. 18] TRANSMISSION OF SIGNALS. ALMOST PLANE WAVES 179
Here, we have put &x d, A If = 2 K L because the inaccuracy k
obtains on both sides of the axis of the beam.
The dimensions of the reflector of the locator itself can be ignored if
the diameter of the beam is considered at a great distance ; and this is
of practical interest. In other words, d is determined only by the
relationship (18.12) and is independent of the dimensions of the re
flector.
The divergence of the beam of rays at every point is measured by
the ratio  . For this reason, the ratio of the cross section of the beam
fc
d to the distance from the locator I cannot be less than the quantity

This relationship is shown in Fig. 28 for the limiting case of the
equality. However, it must be borne in mind that it is not in reality
an equality but an estimate of order of magnitude. ( 1 8. 1 2) is also approx
imate and the symbol > must be written in it.
Thus, we have obtained two estimates for k :
k JL > T (lower estimate)
and, from (18.13),
kd nd K
 =  (upper estimate) . Fig. 28
Eliminating k from these estimates, we obtain
or finally
ltd TT
~7x~ ~ ~d
d>VT\. (18.14)
For example, if Z=100 km and X = l m, then the position of the
object cannot be determined with an accuracy exceeding 320 m.
This is why the dimensions of the reflector could be neglected in the
estimate.
The limit of applicability of the concept of a ray. Equations (18.10)
indicate within what limits the concept of a ray is applicable in optics.
Obviously, one can talk about a ray in a definite direction only when
AJMfc, (18.15)
i.e., when the transverse broadening of the wave vector is considerably
2 TT 2 TT
less than the wave vector itself. But A& ~  r and k~ so that
d x
(18.15) is equivalent to the condition
d>X. (18.16)
12*
180 ELECTRODYNAMICS [Part II
In other words, the dimensions of the region in which the concept
of a light ray is defined must be considerably larger than the wave
length of the light wave. For example, a small circle in the wall of a
cameraobscura of diameter, say, 1 mm is considerably greater than
the wavelength of visible light, which is of an order of magnitude
0.5 x 10~ 4 cm. Therefore, the image obtained in a cameraobscura is
formed with the aid of light rays.
The optics of light rays is called geometrical optics. A ray is defined
only when its direction is given, i.e., the normal to the wave front.
If we are given a beam of nonparallel (for example, converging) rays,
then the wave front is curved. But the radius of its curvature at each
point is considerably greater than the wavelength. Such a converging
beam of rays represents a set of normals to an "almost plane" wave.
The curvature of the wave front close to the focus of the rays may
become comparable with the wavelength, and then there arise devia
tions from geometrical optics. Such deviations are called diffraction
effects. They are also observed when a light wave falls on some opaque
obstacle. In accordance with geometrical optics, we should have ob
tained a sharp shadow a transition from a region where the field differs
from zero to a region where it is equal to zero. But Maxwell's equa
tions do not permit such solutions, which are discontinuous in free
space (cf. the boundary conditions, exercise 1, Sec. 16). In actual fact,
there always exists a transition zone between "light" and "shadow,"
in which the wave amplitude changes in a complicated oscillatory
way.
Exercises
1) Find the limiting dimensions of an object which may be observed in
a microscope using light of wavelength X.
Denoting the semiangle of the cone of rays, drawn from the microscope
objective to the object, by 6, we have Afc = fc sin 0. Whence
Afc A; sin 6 sin0
It is therefore convenient to use a beam of rays with large solid angles
and small wavelengths.
2) Show that if the dispersion law of exercise 19, Sec. 16, is used, then v < c.
We write down an expression for the inverse of v:
I 8k =
v d co
This loads to the inequality
co de
The derivative^ is everywhere positive so that, for s>l, the inequality
can be seen directlv. When e < 1 we have
Sec. 19] THE EMISSION OF ELECTROMAGNETIC WAVES 181
/
. _
2 Sw ~ (w*<o 2 ) 2 f \ m ~
Squaring both sides of the inequality, it is easy to see that this quantity is
greater than I/ 1 H  a_ M a = Ve~.
Sec. 19. The Emission of Electromagnetic Waves
Basic equations and boundary conditions. So far we have considered
electromagnetic waves irrespective of the charges producing them.
In this section we shall consider the emission of waves by point charges
moving in a vacuum. The basic system of equations in this case is
(12.37) and (12.38) together with the Lorentz condition (12.36). We
rewrite these equations anew:
dt*
We begin the solution with (19.2) proceeding in the following man
ner. We assume that p differs from zero only in an infinitesimal vol
ume dV. We find the potential 9 for such a "point" radiator. By virtue
of the linearity of equation (19.2), the potential of the entire spatial
distribution of charge appearing on the righthand side of (19.2) is
equal to the integral of the potentials due to infinitesimal small ele
ments of charge 8e=pdV.
In order to determine the solution uniquely, we must impose a cer
tain boundary condition. It is assumed that the charges are situated
in infinite space, i.e., that there are no conductors or dielectrics any
where.
In free space, a boundary condition can be imposed only at an infinite
distance away from the charges. In accordance with the posed problem
of radiation, it is natural to suppose that there was no field for an
infinitely large interval of time before the initiation of radiation at an
infinitely large distance from the radiator:
9(> oo, r>oo)==0,
(19.4)
A(t > oo, r>oo) = 0.
If no boundary conditions are imposed on the solution of an inhomo
geneous equation, then any solution of the homogeneous equation can
always be added to it so that a unique answer cannot be obtained.
The radiation of a small element of charge. Let us begin with an infinitesi
mal charge element 8e =p dV. We place it at the coordinate origin. Then
182 ELECTRODYNAMICS [Part II
the solution of (19.2) will possess spherical symmetry. In Sec. 11, an ex
pression for the Laplacian operator A was derived in spherical coordi
nates (11.46). As in the case of a static charge [equation (14.7)], we
must retain only the term involving differentiation with respect to r
and, this time, obviously, we must also differentiate with respect to
time. For the time being we consider that the charge density at all
points, except the origin, is equal to zero. Therefore, for all points
for which r^O, equation (19.20) is written thus:
Temporarily, we put
v ^i 11 c\ a\
9 = . (ly.o)
Then
~07 ^ 7" ~0r~ ~~ T 2 " ' r ~07 ==r ~0r~~~'
2 09 2 O 0O 0O 2 <$
0r 0/* 0r 2 0/* 0/* 07* 2 *
Substituting this in (19.5) and multiplying by r (by convention, r is
not equal to zero), we obtain
But this is the equation, of the form (17.5), for the propagation of a
wave. Its solution is similar to (17.12):
The solution ^ depends on the argument t + , and the solution
O 2 depends on the argument t  . The first of these arguments,
t\ , for r>oo, > oo has a completely indeterminate form oo oo,
c
i.e., it is equal to anything. From the condition (19.4), the function <&
becomes zero when r>oo, t> oo. Therefore O x becomes zero for
any value of the argument, i.e., it is equal to zero everywhere. (The
potential at infinity must tend to zero more strongly than so that
there should be no radiation; see below in this section.) For the
function O 2 , condition (19.4) denotes that <D 2 ( oo)=0. In other
words, the function O 2 tends to zero at minus infinity. It does not
follow from this, of course, that it is equal to zero everywhere. Thus,
Omitting the index 2, we write the expression for 9 as follows:
Sec. 19] THE EMISSION OF ELECTROMAGNETIC WAVES 183
9 = 7* () ()
The function <I> is not as yet determined. From the form of its argu
ment we conclude that it describes a travelling wave in the direction
of increasing radii (because >0). Such a wave is termed diverging.
Retarded potential. The value of the function at r = 0, = is shifted
to the point r in a time t = ~ or, in other words, the potential at the
c
point r and time t is determined by the charge, not at the instant of
time t, but at an earlier instant t  . The term  is a measure of the
' c c
retardation occurring as a result of the finite velocity of propagation
of the wave.
But when the retardation becomes a very small quantity, the
c
potential very close to the charge must be determined by the instan
taneous value of the charge de (t). We know from Sec. 14 that the
8&
potential due to a point charge is equal to  (14.8), whence
<b(t) _ 8e(a) _ p (t) d V
r r r
Therefore,
p(i)dF, (19.10)
and the retarded potential of a point charge 9 it I is, in accordance
with (19.9) and (19.10), equal to
JllltrfF. (19.11)
^ '
Now, displacing the coordinate origin to another point, we obtain,
like (14.9),
^_J=Jl) = ML nr 2. T V (IK . (19 . 12)
Here it is assumed that the charge density is given at the point
r (x, y, z), and the potential is calculated at the point R (X, Y, Z),
thus introducing the explicit dependence of p on the spatial argument r.
Finally, in order to obtain the complete solution to (19.2), we must
integrate (19.12) over all the volume elements, i.e., with respect to
a v == ' cLoc dy 0/2 1
9= I Rr ' dV ' (19.13)
For point charges, p denotes the special function that was defined
in Sec. 12.
184 ELECTRODYNAMICS [Part II
Equation (19.1) has exactly the same form as (19.2) and its solution
satisfies the same boundary conditions. Therefore, the vector poten
tial is written quite analogously to (19.13):
ry /in IA\
R_ r  "P. (LJ.U)
Comparing (19.14) with (15.10), which gives A for a stationary
current, we see that j depends on the argument r in two ways: first,
directly, in accordance with its spatial distribution and, secondly,
via the time argument; since the system of currents is not infinitely
small, but has finite dimensions, the retardation of a wave from
various points of the system is different.
Retarded potential at a large distance from a system of charges.
We shall now look for the form of the solutions of (19.12) and (19.13)
at a great distance away from the radiating system. We note that the
integrand depends on the argument R in both integrals in two ways:
in the denominator and via the argument t. The function in the
denominator depends very smoothly on R. Its expansion in terms
of powers of R yields terms which tend to zero like ^ at large dis
tances away from the system. As will be shown later, they do not
add anything to the radiation (for n>l). So we simply replace
l I
TRlITT ^7 ~jf At large distances from the system, the term  R r  ,
appearing in the argument t of the numerator, looks like this:
I I . . ^ , ( . . o)
where n is a unit vector in the direction of R. The subsequent terms
of the expansion (19.15) contain R in the denominator and are insignif
icant. Thus, at a large distance from the radiating system, the poten
tials are:
An estimate of the retardation inside a system. The term  y in
the arguments of the integrands of (19.16) and (19.17) indicates by
how much an electromagnetic wave, coming from the more distant
parts of the radiating system, is retarded in comparison with a wave
radiated by the nearer parts of the system. In other words, the term
determines the time that the electromagnetic wave takes in
passing through the system of charges. If the velocity of the charges
Sec. 19] THE EMISSION OF ELECTROMAGNETIC WAVES 185
is equal to v, then, in that time, they are displaced through a distance
v ^51. The retardation inside the system is negligible when this
c
distance is small in comparison with the size of the system r. Therefore,
if v <^r (or, more simply, v<^.c), then the charges do not have
c
time to change their positions noticeably during the time of propa
gation of the wave in the system.
However, in order that nothing should really change in the system,
the charges must also maintain their velocities in that time, because
the vector potential depends on the currents, i.e., on the particle
velocities. This imposes a further condition which is formulated in the
following manner. Let the charges oscillate and radiate light of fre
quency co. The wavelength of the light is equal to X = . In the time
the phase of the charge oscillations changes by co  ~ . This change
c c
must be small in comparison with 2 TT, whence it follows that the size
of the system must be small compared with the wavelength of the
radiated light in order that the retardation inside the system should
be insignificant. Thus, the term ^51 f n the argument of the integrand
c
is small provided two inequalities are fulfilled: v<^c } r<^X.
The vector potential to a dipole approximation. Let us assume that
both the inequalities obtained have been fulfilled. We omit the term
 in the time argument in the expression for vector potential (19.17).
Then the whole integrand will refer to the same instant of time
t  and we obtain
c
Recall now that j = pv and that the charges are point charges.
Then the integral (19.18) is reduced to a summation over separate
charges :
T>
Here, the lower index t  denotes that the whole sum must be
c dr j
taken at that instant of time. But v 1 ' y, so that
Here we have used the definition for dipole moment (14.20). We
note that (19.20) involves only a time derivative d. Therefore, the
186 ELECTRODYNAMICS [Part II
transformation (14.21), which corresponds to a change of coordinate
origin, does not change d either for a charged system or for a neutral
system. In particular, (19.20) holds also for a single charge.
The approximation (19.20), in which A is expressed in terms of a
derivative of the dipole moment of the system as a whole, is termed a
dipole approximation.
The Lorentz condition to a dipole approximation. In Sec. 17, a poten
tial gauge transformation for travelling plane waves was chosen such
that the scalar potential became zero. We shall make the same gauge
transformation for diverging spherical waves. To do this we must
subject the vector potential to the following condition:
divA = 0, (19.21)
which is obtained from (19.3) if we take 9 = 0.
In condition (19.21) we should not differentiate A with respect to
R in the denominator: each such differentiation increases the degree
of R by unity, while the potential is determined at a large distance
from the radiating system. Only terms inversely proportional to R
n
contribute to the radiated energy (see below). The unit vector n = p,
which will appear in differentiation, need not be differentiated a second
time, since that would also give rise to superfluous degrees of R in the
denominator.
We choose an arbitrary gauge function in the form
_ I R \
n n * /  , i
11 UL \ v ^ 
/=  '  . (19.22)
Then, applying equation (11.37), it is easy to see that the condition
(19.21) is fulfilled:
1 1 f
And the scalar potential is cancelled by  ~.
c (/ 1
Field to a dipole approximation. Let us now calculate the electro
magnetic field. We need to differentiate only inside the argument
t  . In calculating the magnetic field, we make use of (11.38) and
of the fact that rot grad / = 0:
H = rot A = i rot d
The electric field is
1 0A 10 _. d , 1 , 3.
T " ~ V f =   + * (nd) =
(19.24)
Sec. 19]
THE EMISSION OF ELECTROMAGNETIC WAVES
187
From these equations it can be seen that the electric field, the
magnetic field, and the vector n are mutually perpendicular. In addi
tion, H = E, since E*= [Hn] 2 // 2 (Hn) 2 , and Hn = 0, since H and
ii are perpendicular. Consequently, the wave at a point R at a great
distance away from a radiating system is of the nature of a plane
electromagnetic wave. This result was to be expected because the
field is calculated far away from charges, where the wave front may
be approximately regarded as plane and the solution becomes the
same as obtained in Sec. 17.
Fig. 29 gives a general picture of the field. We situate the vector d
at the centre of a sphere of large radius R so that d coincides with the
polar axis or, in other words, is directed towards the "north pole."
Let us draw the radius vector of some point. Through this point, we
draw the meridian and the parallel. Then the electric field is tangential
to the meridian and is directed "towards
the south," while the magnetic field is
tangential to the parallel and is directed
"towards the east." It can be seen from
equations (19.23) and (19.24) that the
field becomes zero at the poles and max
imum 011 the equator, i.e., on a plane
perpendicular to d. The field distribu
tion in space does not possess spherical
symmetry. We note that the trans
verse field cannot be spherically sym
metrical for purely geometrical reasons.
The zone, in which the field is calculat
ed according to equations (19.23) and
(19.24), is called a wave zone.
The intensity of dipole radiation. Let us now find the energy lost
by the system in radiation. We must calculate the energy flux crossing
an infinitely distant surface. The energy flux density, or the Poynting
vector, is
(19.25)
Fig. 29
Hence, the energy flux is directed along a radius, as it should be in a
wave zone. The total energy crossing a sphere of radius R in unit time is
dt
because the vector ds is directed along n. Further,
where is the polar angle. From (19.23)
(19.26)
TU sin8d,
(19.27)
188 ELECTRODYNAMICS [Part II
By substituting (19.27) in (19.26), cancelling R 2 , and integrating, we
obtain an expression for the energy radiated in one second:
.
dt ~ 3 ~& ' .
We note that all the terms in the expression for fields containing R
in the denominator to a higher degree than the first would not con
tribute anything to (19.28) for a sufficiently large R. It is for this reason
that only first degree terms in J?have been retained in the denominator.
The significance of equation (19.28). Equation (19.28) expresses a
result of fundamental importance energy is radiated whenever a
charge is accelerated. Indeed, d = JT'e, f ' . Hence it is necessary so
that d should differ from zero that the charges should be in accel
erated motion irrespective of the sign of the accelerations.
But then electrons moving in an atom should radiate energy con
tinuously and should fall into the nucleus ; every electron is in accelera
tion motion, otherwise its motion could not be finite (see Sec. 5).
In actual fact atoms are obviously stable and the electrons do not
fall into the nucleus.
Here, we realize that classical, Newtonian, mechanics can in 110
way be applied to the motion of an electron in an atom. In the third
part of this book we shall explain the stability of atoms using quantum
mechanics, where the very concept of motion differs qualitatively
from that in classical mechanics.
Magnetic dipole and quadrupole radiation. We have indicated that
charges must be in accelerated motion to radiate. But a simple example
can be given when equation (19.28) yields zero even for accelerated
charges. Let the system consist of two identical charged particles.
According to Newton's Third Law their accelerations are equal and
opposite in sign, so that d = &* ' = ^(r 1 + f 2 )= : 0. In this case
i
this law is applicable because, to a dipole approximation, the retar
dation of electromagnetic interactions inside the system is considered
as negligibly small and, hence, the interaction forces between charges
are regarded as instantaneous. But there is then no need to take
account of the momentum transmitted to the field and the total mo
mentum of the particles is conserved, thereby leading to the condition
?!= r 2 . For this case, approximation (19.18) does not describe the
radiation and it becomes necessary to use higherorder approximations.
If, in the expansion in powers of , a further term is retained in
c
addition to the zero term, then a radiation is obtained which depends
on the change in magnetic quadrupole and dipole moments of a system
of charges. It is essential that this expansion be not in terms of inverse
Sec. 19] THE EMISSION OF ELECTROMAGNETIC WAVES 189
powers of R, as in electrostatics and magnetostatics, but in powers
of the retardation inside the system.
We have already mentioned that the retardation inside the system
is small when v <^ c and r <^ X. The ratio is involved in the
c
magnetic moment of the system. Therefore, those terms in the ex
pansion (involving powers of the retardation) which are proportional
to  ~ , account for magnetic dipole radiation. The quadrupole moment
of the system involves an additional power of r compared with the
dipole moment, and so quadrupole radiation is related to those terms
of the expansion which are proportional to ~ . Higher approximations
A
are important in those cases for which lowerorder approximations,
for some reason or other, become zero, as is the case of the two identical
charges.
The field due to a magnetic dipole radiator is similar to the field
of a radiating electric dipole. Unlike the field represented in Fig. 29,
the magnetic field, for magnetic dipole radiation, lies in the plane jx,
(i.e., it is along a meridian), while the electric field is along a parallel.
The equation for intensity is similar to (19.28), though it involves
(ti) 2 instead of (d) 2 . Since the magnetic moment is proportional to ,
c
the intensity of magnetic dipole radiation is less than the intensity of
electric dipole radiation in the ratio I 1 .
The field of a radiating electric quadrupole has a more complicated
configuration. The expression for the intensity of such radiation
involves the square of the third derivative of the quadrupole moment
of the system. In order of magnitude, the intensity of quadrupole
(r \2
I .
Exercises
1) Calculate the time that it takes a charge, moving in a circular orbit
around a centre of attraction, to fall into the centre as a result of the radiation
of electromagnetic waves. Regard the path as always approximately circular.
2) A particle with charge e and mass m passes, with velocity v, a fixed
particle of charge e 19 at a distance p. Ignoring the distortion in the orbit of
the oncoming particle, calculate the energy that this particle loses in radiation.
00 00
A AJP 2 &2 fr12.7, 2 e*ef f dt w e*f
Answer: A^ , d  A J (p> + ^ i)a 1.
3) Why is it that when two identical particles collide (e 1 =e 2 , m 1 = m 2 ),
magnetic dipole radiation does not result if the interaction is calculated according
2 LL 2
to the Coulomb law ? The intensity of magnetic dipole radiation is   .
190 ELECTRODYNAMICS [Part II
4) A plane light wave falls on a free electron causing it to oscillate. The
electron begins to radiate secondary waves, i.e., it scatters the radiation.
Find the effective scattering cross section, defined as the ratio of the energy
scattered in unit time to the flux density of the incident radiation.
We proceed from the fact that r = and then determine y from (19. 28).
Dividing by the energy flux , we obtain
Sec. 20. The Theory ol Relativity
The law of addition of velocities and electrodynamics. In Sec. 15,
the interactions of charges with a magnetic and an electric field was
reviewed [see (15.34)]. But the motion of the charges was considered
slow, in other words, the velocity satisfied the inequality v <^ c.
Yet this inequality is by no means always satisfied. Electrons ob
tained in beta decay, particles in cosmic rays, and particles in accel
erators move with velocities close to that of light. Hence, it is necessary
to obtain the laws of mechanics for these ultrahighspeed charged
particles.
If we attempt to apply Newtonian mechanical laws to these particles
we will encounter an insurmountable contradiction the law of
addition of velocities cannot be applied in electrodynamics in its
usual form (see Sees. 8 and 10).
The equations of Newtonian mechanics are of the same form for
all inertial systems moving uniformly relative to each other. In such
systems there are no inertial forces. Naturally, under no circumstances
can the principle of the equivalence of inertial systems be violated,
otherwise we should have to assume that there existed a reference
system at absolute rest. We must also consider that the equations of
electrodynamics appear the same for all inertial systems in free space
in the form (12.24)(12.27). It follows from these equations that the
velocity of propagation of electromagnetic disturbances is equal to c
and is the same for all directions in space. If it turned out that in
some inertial systems the velocity of light depended upon the direction
of its propagation, then these systems would not be equivalent to a
system in which the velocity of light is the same in all directions. In
this system, the electrodynamical equations would admit of a solution
in the form of a spherical wave a solution similar to the one that was
obtained in the preceding section. In all other inertial systems, the
velocity of light would depend on the direction of the wave normal.
An analogy would arise with the propagation of sound in air : the
velocity of sound in a system at rest relative to the air does not depend
on direction, but in a system moving relative to the air the velocity
Sec. 20] THE THEORY OF RELATIVITY 191
of sound is less in the direction of motion and more in the opposite
direction as a consequence of the law of velocity addition.
So far, it has been considered that light is transmitted in an elastic
medium, "the ether," and it has been regarded as selfevident that the
velocity of light must be governed by the same law of velocity addition
as the velocity of sound in air. Then a reference system fixed in the
"ether" would have to be regarded as being at absolute rest, while all
the remaining systems, as in absolute motion. In these systems the
velocity of light would depend on its direction of propagation, in
accordance with the law
c'=c + v, (20.1)
where, for simplicity, only that direction is taken which coincides with
the relative velocity of the system.
Michelson's experiment. A direct experiment was performed which
showed that the velocity of light cannot be combined with any other
velocity and, in all reference systems, it is equal
to a universal constant c. This was the famous
Michelson experiment (1887) which we shall
describe in brief. ^A ray of light falls on a
halfsilvered mirrorS/S (Fig. 30), where it is
split up : a part of the light is reflected and falls ^
on mirror A while the other part is transmitted '
and falls on mirror B. Let SA be perpendicular *
to the motion of the earth and let SB be
parallel to the earth's motion. The light reflect
ed from the mirrors A and B returns to the
plate SS; the ray BS is reflected from it and ^
falls on screen C while the ray AS is trans _ ^u
mitted to the screen directly. Thus, both rays Fig. 30
are entirely equivalent as regards transmissions
and reflections, though in the sections AS and BS the light is propa
gated differently relative to the earth's motion.
Let us assume now that the velocity of light is combined with the
motion of the earth according to the usual law of addition of veloc
ities. Then, along the path SB, the velocity of light relative to the
earth is equal to c V, and c+V along the return path, where V
is the velocity of the earth. The time light takes to travel along the
entire path SBS in both directions is
I I 2lc 21 21V Z
where I SB. We have used the fact that F<^c. Along the section
SA the velocity of light and the earth's velocity are perpendicular to
each other (in a reference system fixed in the apparatus). If again the
law of velocity addition holds, then the velocity of light relative to
192 ELECTRODYNAMICS [Part II
the apparatus, along the section SA, is equal to Vc 2 F 2 (c is the
hypotenuse of the triangle, F and Vc 2 F 2 are the sides). The time
taken by the light to travel along the whole path SAS, equal to 2 Z, is
21 _ 21 IV 2
C 3
Thus, the difference in the passage times along the paths 8B8
and SAS is equal to y . By means of repeated reflections, the path
can be made sufficiently long (several tens of metres). Choosing it
properly, we can arrange that the proposed time difference for the paths
SA S and S BS is equal to a halfperiod of oscillation. The rays on the
screen C should then mutually cancel. In order to be certain that the
cancellation has occurred as a result of the combination of the velocity
of light with the velocity of the earth, and not by accident, it is suffi
cient to rotate the apparatus through 45 so that the direction of the
earth's velocity would be along the bisector of the angle A SB. Then
the difference in time taken by the rays to travel along the paths
SAS and SBS should at least become equal to zero; hence, if, in the
previous position, the rays mutually cancelled as a result of the com
bination of the velocity of light with the earth's velocity then, in the
new position, the rays would mutually reinforce each other. In other
words, the interference bands on the screen would be displaced by
half a wavelength.
Actually, no change in the path difference between the rays occurs
when the apparatus is rotated, i.e., the expected effect is completely
absent. An addition of the velocity of light and the earth's velocity
does not occur.
The negative result of Michelson's experiment is completely under
standable if we reject the postulate of an "ether." Nevertheless,
at the time when the experiment was performed, no one as yet under
stood that electrodynamics does not require an "ether" in order to
become as complete and clear a science as mechanics. The fact that
the law of addition of velocities a truism for physicists in the past
failed to hold, appeared as an inexplicable paradox.
In addition, Michelson's experiment apparently contradicted the
phenomenon of astronomical abberation of light and Fizeau's experi
ment. (These will be considered later in this section in the light of the
theory of relativity.)
Einstein's relativity principle. We shall not give the history of the
painful attempts to explain this paradox but, instead, we will straight
way present the correct solution to the problem as given by A. Ein
stein in 1905. It is known as the special theory of relativity. Despite the
delusion widespread among laymen, this term by no means expresses
the relativity of our physical knowledge. It expresses the mutual
equivalence of all inertial systems moving relative to one another.
Sec. 20] THE THEORY OF RELATIVITY 193
The equations of electrodynamics do not imply the presence of any
elastic medium ("ether") for the transmission of electromagnetic
disturbances. We have already discussed this in Sec. 12. The reality
is the electric field itself. For this reason, the equations of electro
dynamics are just as independent of the choice of inertial reference
system as the equations of mechanics. Both sets of equations describe
motion, i. e., the change of state with time directly. Mechanics de
scribes the change of mass configuration, while electrodynamics de
scribes changes of the electromagnetic field. The forms of the equations
of motion cannot change as a result of the choice of inertial system.
This is why the result of Michelson's experiment does not contradict
the notion of relativity of motion, but confirms it. Michelson's experi
ment shows that the velocity of light in free space is the same in all
inertial systems. The velocity of propagation of interactions is a funda
mental constant in the equations of electrodynamics. These equations
are invariant to a transformation from one inertial system to another
only when the velocity of propagation of the interactions in both sys
tems is the same. And so the result of Michelson's experiment contra
dicts only the law of addition of velocities, i. e., Galilean transforma
tions (8.1) and (8.2). This law of addition of velocities is confirmed
experimentally only for relative velocities and for velocities of motion
that are small compared with the velocity of light c. Obviously, it
must be replaced by a more precise law for the region of high velocities.
But this more precise law must also hold in mechanics for large par
ticle velocities. This can be seen from the following reasoning.
Let the charges in a specific inertial system interact in some way
with the electromagnetic field producing certain events (for example,
collisions between the charges). They may be precalculated on the
basis of the equations of mechanics and electrodynamics. In trans
forming to another inertial system, the equations of mechanics and
electrodynamics must retain their form, otherwise, other consequences
will follow from the transformed equations taken together ; in partic
ular, those events which were precalculatecl and occur in the first
inertial system do not necessarily take place in another system.
But events such as collisions, for example, are objective facts; they
should be observed in all coordinate systems. Yet if we apply Galilean
transformations (8.1) then the equations of Newtonian mechanics will
not change, while the equations of electrodynamics will change, since
the law of addition of velocities (10.12) is not applicable in electro
dynamics. Therefore, we must find transformations to replace the
Galilean transformations such as would leave both the equations of
mechanics and the equations of electrodynamics invariant. But then
it would become necessary to make the laws of Newtonian mechanics
more precise, since they are correct only for low particle velocities.
A physical theory which cannot predict facts independently of the
mode of their description is imperfect and contradictory. It is this
13  0060
194 ELECTRODYNAMICS [Part II
that makes us reconsider the basic facts of mechanics, no matter how
selfevident they seem to be in our everyday experience, which has
to do with the motions of bodies at velocities that are small compared
with that of light.
The Lorentz transformations. We look for transformations of a more
general form than, the Galilean transformations for passing from one
inertial system to another. Like the Galilean transformations, they
must satisfy certain requirements of a general nature. These require
ments may be expressed as follows.
1) The transformation equations are symmetrical with respect to
both systems. We shall denote the quantities that refer to one system
by letters without primes (x, y, z, t), while those that refer to the other
system will be primed (#', y', z', t'). We denote the velocity of the primed
system with respect to the unprimed system by F. Then the mathe
matical form of the equations expressing unprimed quantities in
terms of primed quantities (and the velocity F) is the same as that
of the equations for the reverse transformation, if we change the sign
of the velocity in them. This requirement is necessary for the equiva
lence of both systems.
2) The transformation must convert the finite points of one system
to the finite points of the other, i.e., if (x 9 y, z, t) are finite, then a
transformation with finite coefficients must leave (x r , y', z', t') finite
values.
Condition (1) greatly restricts the possible form of the transforma
tions. For example, it can be seen that the transformation functions
cannot be quadratic, because the inversion of a quadratic function
leads to irrationality, just as that of the function of any degree other
than the first. A linearfractional transformation (i.e., the quotient of
two linear expressions) under certain limitations imposed on the
coefficients may be inverted retaining the same form. For example,
for one variable the direct and the inverse linearfractional functions
look like
/ ax}b b fx'
ex} f ' ~~ ex' a
But this function does not satisfy condition (2): if x' =a/e, x be
comes infinite. Therefore, a linear function is the only possible one.
3) When the relative velocity of two systems tends to zero, the
transformation equations yield an identity (x' =x, y' =y, z' =2, t' =t).
4) A law of the addition of velocities is obtained from the transfor
mation equations such that it leaves the velocity of light in free space
invariant: c'=c.
Summarizing, we can say that the transformation equations: 1)
maintain their form when inverted, 2) are linear, 3) become identities
for small relative velocities, 4) leave the velocity of light in free space
unchanged.
Sec. 20] THE THEORY OF BELAT1VITY 195
These four conditions are sufficient. The required equations can be
obtained most simply if one of the coordinate axes (for example, the
#axis) is taken in the direction of the relative velocity. Then the other
axes will not be affected by the transformation.
We return to Fig. 11 (page 68), but we will not make the arbitrary
assumption that t = t' (experiment supports this only for small relative
velocities of both systems). Let us see what results from the conditions
(l)(4). If the velocity is along the #axis, then, as has just been found,
y' = y, z' z. This can be seen simply from Fig. 11. In the most general
form, linear transformations of x and t, appear thus :
x' = ax + $t , (20.2)
t' = y# + St. (20.3)
The constant terms need not be written in these equations ; they can
be included in the definition of x or x' through choice of the coordinate
origin.
Let us apply equation (20.2) to the origin of the primed system,
x' = 0. This point moves with velocity V relative to the unprimed sys
tem. Hence, xVt. Substituting re'=^0, x=Vt in (20.2), we obtain,
after eliminating t,
<xFhP = 0. (20.4)
We shall solve equations (20.2) and (20.3) with respect to x and t.
Elementary algebraic computations give
x= ,^=^_ (20 . 5)
aS PY
'= (20 ' 6)
Let us now apply condition (1). For this we note that the coefficients
(5 and y, which interrelate the coordinate and time, must change
sign together with the velocity V. Otherwise, if the x and x r axes
are turned in the opposite direction the equations will not preserve
their form, and this is impermissible. Thus, the equations for the
inverse transformation from unprimed quantities to primed have
the same form as (20.2) and (20.3):
x  art  $t' , (20.7)
t = ya;' + S*' . (20.8)
Comparing (20.7) and (20.5), we obtain
= T8=TT . (209)
P =  a T_V < 2<UO >
13*
196 ELECTRODYNAMICS [Part II
From (20.10) it follows that
a8 py=l. (20.11)
Then, from (20.9), we obtain
<x 8. (20.12)
No other relationships are obtained from the comparison of the
direct equations with the inverse equations.
We now use condition (4). We divide equation (20.2) by (20.3):
( 20  13 )
Let # be a point occupied by a light signal emitted from the origin
of the unprimed system at an initial instant of time = 0. Obviously,
~ =c. But in accordance with condition (4), 7 =c. Hence,
t v
.
yc t 8
(20.14)
v '
We substitute the relations (20.4) and (20.12) into (20.14) in order
to eliminate p and 8. There remains a relation between a and y:
yc 2 + ac = ac aF ,
whence
YOCJ. (20.15)
We now substitute (20.15), (20.4) and (20.12) into (20.11) and obtain
an equation for a:
(20.16)
In extracting the square root, we must take the positive sign
in accordance with condition (3), because then (20.3) becomes t 1 t
for a small relative velocity. A minus sign would yield t' = t,
which is meaningless.
Now expressing all the coefficients a, (3, y, and 8 in accordance
with equations (20.16), (20.4), (20.15), and (20.12), respectively, and
substituting (20.3) into (20.2), we arrive at the required trans
formations :
X ' = *~ t , (20.17)
(20.18)
Sec. 20] THE THEORY OF RELATIVITY 197
These equations are called Lorentz transformations. From (20.7)
and (20.8), the inverse transformations are of the form
(20.19)
(20.20)
In order to explain the meaning of these equations we shall apply
them to some special cases. Let a clock be situated at the origin
x' =0 of the primed system. It indicates a time t'. Then, from equation
(20.20), it follows that
The clock which is at rest relative to its reference system we call
the observer's dock. It can be seen from (20.21) that one observer,
comparing his clock with that of another observer, will always observe
that the latter clock is slow, i.e., that t'<^t. If a clock is situated
at the origin of the unprimed system (i.e., at the point # = 0), the
transformation equation to the primed system is of the same form,
since, from (20.18), we now obtain t' = t . This not only
does not contradict (20.21), but expresses that very fact: a clock
moving relative to an observer is slow compared with his own clock.
In the theory of relativity, a single universal time does not exist
as in Newtonian mechanics. It is better to say that the absolute
time of Newtonian mechanics is, in actual fact, an approximation,
correct only for small relative velocities between clocks. The absolute
ness of Newtonian time has sometimes given cause to regard it as
an a priori, logical category independent of moving matter.
At any rate, Newton, by accepting instantaneous action at a
distance, naturally had to consider time as universal ; if we formally
put c = oo in (20.18), we obtain t' =t. The instantaneous transmission
of signals would allow us to synchronize clocks in all inertial systems
independently of their relative velocities. In Newtonian mechanics,
gravitational forces played the part of such instantaneous signals.
It is sometimes thought that, knowing the velocity of light c,
we can introduce a correction into the readings of clocks in different
inertial systems such that the rate of time will everywhere be the
same. But it is precisely equation (20.21) that describes the relative
198 ELECTRODYNAMICS [Part II
passage of time in both reference systems after a correction has been
introduced for the finite time of propagation of light. Time reduction,
as has already been shown, is completely reciprocal. Consequently,
it can in no way be accounted for by any change, resulting from
motion, in the properties of clocks. The time reduction effect is purely
kinematical.
It must also be added that in speaking about clocks we by no means
necessarily have in mind clocks which have been made by human
hands; any natural periodic process that gives a natural time scale
will do as well, for example, the oscillations in a light wave. It is
clear that the physical properties of a radiating atom cannot in the
least depend upon the inertial system in which the atom is described.
This is what gives us the right to assert that equation (20.21) refers
to similar clocks.
At the same time we must remember that it is impossible to define
time without relation to some periodic process, i.e., irrespective of
motion.
Relativity and objectivity. The relativity of time by 110 means
indicates a rejection of the objectivity of its measurement in any
given system of reference. It is entirely of no consequence what
observer is observing the clock. The relative character of time in
inertial systems is the only thing that counts. We have all long since
become accustomed to the relativity of the datum line of time measure
ment related to time zones (i.e., to the sphericity of the earth). The
theory of relativity teaches us that the time scale is also relative.
The fact that an objective concept may be relative can be seen
from the following example. In the Middle Ages it was thought
that even direction in space was absolute, and it was then thought
impossible to imagine that the earth was spherical since it would
then follow that our antipodes would have to walk upside down!
The concept of "up" or "down" was related not with the direction
of a plumbline at a given point on the globe, but with certain other
categories characteristic of the ideology of the Middle Ages. The
vertical directions in Moscow or Vladivostok form a substantial
angle between each other, but nobody nowadays would think of
arguing about which of them is the more "vertical." The concept
of verticality is completely objective at every point on the globe,
but is relative for different points. In the same way, time is objective
in each inertial system, but is relative between them.
Contraction of the length scale. We shall now consider the question
of the measurement of length. In order to find out the length of
a moving body ("its scale"), we must simultaneously plot the co
ordinates of its ends in a fixed system. Obviously, a fixed observer
has no fundamentally different means of measurement, for, otherwise,
he would have to stop the motion of the scale (i.e., transfer it to
his reference system). If the ends of the scale are fixed by a stationary
Sec. 20] THE THEORY OF RELATIVITY 199
observer at one time* we must put = 0. From (20.17), there follows
an expression for the length of a moving scale Aa;' measured by a
fixed observer:
A*' = n ** . (20.22)
Like (20.21), this equation has a symmetrical inversion. If a
Amoving" observer measures a "stationary" scale, we must put
V ==0; it then turns out that Lx =   ~ x _ . We conclude from
V'?'
(20.22) that a moving scale is shortened relative to a stationary
observer. The contraction occurs in the direction of motion.
Lorentz supposed that this scale compression does not appear to
both inertial systems, but, for some unknown reason, occurs when
the scale moves relative to the "ether." Lorentz and others attempted
in this way to explain the negative result of Michelson's experiment.
Yet the very symmetry of the direct and inverse Lorentz trans
formations (20.17)(20.20) (they were known before the advent of
the theory of relativity), from which the contraction of length follows
only as a special case, shows convincingly that there is no system
at absolute rest relative to an "ether." It may be noted that by the
beginning of the twentieth century, the "ether," which had been
introduced by Huygens as a medium that transmitted light oscillations,
remained in physics simply as a rudimentary concept. The discovery
and confirmation of the electromagnetic nature of light made the
hypothetical elastic medium quite superfluous (see Sec. 12). Only
the theory of relativity disclosed the real meaning of the Lorentz
transformations. But then, many concepts regarded as absolute in
Newtonian mechanics, turned out to be related to the motion of
inertial systems.
The formula lor addition of velocities. We shall now find an equation
for the addition of velocities arising from the Lorentz transformations.
Differentiating (20.17) and (20.18) and dividing one by the other,
we obtain
.*L_ F
*gr^vl= dt v .  */ . (20.23)
dt x V dx Vv x v '
~~ 1 ~""
* The idea of the simultaneity of two operations performed in the same
coordinate system may be uniquely defined with the aid of light signals. Indeed,
observers at rest relative to each other at a given distance can always check
their time with the aid of light signals by introducing a constant correction
for the known time of propagation.
200 BLECTBODYNAMICS [Part II
Noting that dy' =dy and dz' =dz, we have a transformation of
the velocity components perpendicular to V:
dy' , dt V * c 8 _ y y c* ..., " * * ,2024)
dF s ^ = F35  v^r"' Vz I F5T~' (20 ' 24)
* :
For small velocities, (20.23) and (20.24) become the ordinary
equations for addition of velocities. This can be seen if we let c tend
to infinity, i.e., by putting =0.
It is easy to see that if v = Vv* + v$ + v\ = c , then likewise
v' =c, i.e., the absolute value of the velocity of light does not change
in passing from one inertial system to another. But the separate
components of the velocity of light, which are less than c, may of
course change ; the direction of a light ray relative to different observers
differs, since there is 110 absolute direction in space.
Alteration of light. In this connection let us consider
the phenomenon of the aberration of light. Astronomical
aberration, or the deflection of light, consists in the
\ fact that stars describe ellipses in the sky in the course
\ of a year. Their origin is easy to explain : the velocity
\ of the earth, in annual motion, combines differently
\ with the velocity of the light emitted by the star
\ (Fig. 31). If the velocity vector of the starlight relative
\ to the sun is ES then the resultant direction of the
5 velocity, for one position of the earth, is ET l and, in
Fig. 31 half a year's time, ET 2 . These directions are projected
on different points of the celestial sphere so that in
the course of a year a star describes a closed ellipse. In angular
units, the semi major axis of the ellipse is always equal to
V 1
where V is the velocity of the earth. = 1QOQO = 20". 25.
We may ask the question: Why does not the velocity of light
in Michelson's experiment combine with the earth's velocity but
remains equal to c, while the phenomenon of aberration shows that
velocities combine (we note that Michelson's experiment was also
performed with an extraterrestrial source of light). The explanation
is that in Michelson's experiment it was the absolute value of the
velocity of light c that was measured (from the path difference of
the rays), while in the aberration of light there is a change in the
direction of the velocity of light as a result of the combination of
its components with the velocity of the earth. Considering that the
velocity of light relative to the sun is perpendicular to the plane
Sec. 20] THE THEORY OF RELATIVITY 201
of the earth's orbit, we must put v* = 0, v y c, v z = Q into (20.23)
and (20.24). Then the components of the velocity of light relative
to the earth are
And, in accordance with Michelson's experiment, v' x 2 {v'y 2 =c 2 . The
direction of the projection of the velocity of light onto the plane
of the earth's orbit (ecliptic) is reversed in the course of half a year,
which is the reason why aberration occurs.
Similar equations are obtained in the more complicated case when
the rays from the star are not perpendicular to the plane of the
ecliptic. They coincide with the equations that follow from a simple
F 3
addition of velocities if terms of the order = are neglected.
c
Before the theory of relativity was put forward, it was wrongly
supposed that the aberration of light contradicted Michelson's experi
ment.
Fizeau's experiment. Fizeau's experiment, which determined the
velocity of light in a moving medium, was also believed to contradict
Michelson's experiment. Fizeau's method was this. A beam of light
was divided into two parts using a halfsilvered mirror (Fig. 32).
These beams were passed through tubes with flowing water; one
beam in the direction of flow and the other in the opposite direction.
For comparison, the same beams were passed through tubes in which
the water was at rest. By subsequent reflections the beams once
again combined and cancelled each other when the path difference
between them was equal to an integral number of half wavelengths
(i.e., when they were in opposite
phase). Coherence between them was
obtained due to the fact that they
both came from the same source. In
still water, the path difference was chosen
so that the rays were reinforced, i.e., the ~*
phase difference was equal to an even
number of half wavelengths. The path Fig. 32
difference in flowing water was varied.
Since the frequency of the light and the tube lengths remained un
changed, the change in path length indicated a change in the velocity
of light relative to the tubes.
First of all, we note that the result of Fizeau's experiment in no
way contradicts the general ideas about the relativity of motion.
A reference system fixed in flowing water is not equivalent to a
system fixed in the tube, if we are studying the propagation of light
in water.
202 ELECTRODYNAMICS [Part II
Since the velocity of light in water is equal to , where v is the re
fractive index of the water, the general equation for the addition
of velocities (20.23) shows that does not remain a constant quantity
when passing to another coordinate system. At the same time, we
cannot use the simple velocityaddition equation, because the de
nominator of equation (20.23) differs from unity by (F is the veloc
ity of the water). Considering that V<^c and expanding the de
nominator in a series up to the linear term inclusive, we find the change
in the velocity of light in moving water (see exercise 1):
It was precisely this value, which differs from that given by the
simple velocityaddition law, that was obtained by Fizeau. Since
Michelson measured c and Fizeau measured, there is no contra
diction between them.
Interval. Despite the fact that x and t are changed separately by
the Lorentz transformations we can construct a quantity which
remains invariant (unchanged). It is easy to verify that this property
is possessed by the difference c 2 t 2 x 2 . Indeed,
T72~,'2
c 2 * /2 +  ? + 2Vx't'
 c ,,  ,
x' z + V*t' z + 2 Vx't'
X* =  a
or
c*t* x 2 ^ c 2 *' 2 x' 2 = s 2 . (20.25)
The quantity s is called the interval between two events; that
which occurred at the coordinate origin x = Q and initial time = 0,
and another event that occurred at the point x and time t.
The word "event" may also be regarded in its most common
everyday sense provided that its coordinates and time may be defined.
If the first event is not related to the origin of the coordinate system
and the initial instant, then
^  c 2 (* t  tj*  (x 2  xj 2  c 2 (*;  t[) 2  (x' 2  x[) 2 . (20.26)
Sec. 20]
THE THEORY OF BELATIVITY
203
Considerable importance is attached to the interval between two
infinitely close events:
 dx* .
(20.27)
' A* i
Absolute
future
It is not at all necessary to consider that both events occurred on
the abscissa. Since dy'^dy and dz' = dz, the interval is always in
variant :
ds*  c*dt*  dx*  dy*  dz* = c*dt*  dl* = c*dt'*  dl'* . (20.28)
The interval, written in the form (20.28), is not related to any definite
direction of velocity.
Space and time intervals. The interval provides for a very vivid
way of studying various possible space time
relationships between two events. Let the spatial
distance between the points at which the events
occurred be taken along the abscissa, and the
interval of time between them, along the ordinate
axis (Fig. 33). To begin with, let ct > I , for example,
$2 = C 22 I* (20.29)
We shall plot the values of ct and I corresponding
to two definite events measured in quite different
inertial systems. No matter what values are ob
tained as a result of these measurements of ct and
Z, the interval s (20.29) between the events is the
same. It follows that the locus of the points, for
all possible spatial distances I and time intervals Fig. 33
ct, is an equilateral hyperbola s 2 ^c 2 t 2 I 2 . Two
branches of the hyperbola are possible : one lies in the past relative
to the event that occurred at = 0, #^0, while the other is completely
in the future. It is easy to see that such a relationship inevitably
results if the events are causally related. Let the events, in some
reference system, be known to occur in the same place, for example,
sowing and reaping. To this system there corresponds the point
(sowing) and the point A (reaping). But since all points of the given
branch of the hyperbola lie at >0, the sowing in any reference
system must occur earlier than reaping.
We can also proceed from causally related events which in our
coordinate system do not occur at one point of space, such as firing
and hitting the target (they occur at one point in a system fixed
in the bullet). To the system fixed in the bullet there now corresponds
a vertical section OA in Fig. 33, while to our system there corresponds
some inclined line drawn from the origin to a point on the same
upper hyperbola. Thus, here too, the second event that of hitting
the target occurs after the shot in any reference system.
204 ELECTRODYNAMICS [Part II
If the velocity of the bullet (or any material particle) is v, then
02._. c 2g2 J2 > o. it necessarily follows from the inequality (20.29)
that l = vt<d or v<c, if a reference system exists in which both
events occurred at a single point in space. For this reason, the velocity
of any material particle can only be less than the velocity of light c.
The region above the first asymptote is called "absolute future"
relative to the initial event.
Although the example of a bullet hitting a target may appear
to be a special case, in actual fact the foregoing reasoning may be
used for all cases when an effect is related to a cause by some material
transfer process, either in the form of a particle or in the form of
a wave packet. A reference system may be related to moving matter,
and therefore any velocity of material transport satisfies the inequality
Consequently, the theory of relativity never contradicts the objective
nature of causality. And it is precisely the sequence of cause and
effect which determines the direction of time.
We can also consider other pairs of events. For example, let one
event occur on the sun and the other, five minutes later, on the
earth. Light travels from the sun to the earth in eight minutes;
for the two given events ct<l, s 2 =c 2 t 2 1 2 <Q. For purely physical
reasons, such events can in no way be related causally because the
velocity of transport of matter does not exceed c. For such events,
the interval term Z 2 is not equal to zero in any system of reference.
And so there is no coordinate system, relative to which events related
by an imaginary (s 2 <0) interval occur at a single point of space.
Yet, on the other hand, their time sequence has not been defined:
coordinate systems exist in which the first event occurs before the
second, and there are systems in which the second event occurs
before the first.* Thus, the theory of relativity denies the absolute
nature of the simultaneity of two events occurring at different points
in space and separated by an imaginary interval. and B in Fig. 33
are such events. B lies on the hyperbola which, relative to O, belongs
partly to the future and partly to the past. But and B can in no
way be related causally, since no interaction can arrive at B from O
instantaneously.
Hence, the relativity of simultaneity does not contradict the absolute
nature of causality.
The region between the asymptotes is called absolutely distant,
relative to the coordinate origin.
The light cone. The asymptotes to the hyperbola l=ct are of special
interest. For them, s = 0.
* For example, the second event occurred earlier than the first relative
to any system moving in the direction from the earth to the sun with a velocity
exceeding c, as may easily be seen from (21.20).
o
Sec. 20] THE THEORY OF RELATIVITY 205
The relationship l = ct holds for two events related by an electro
magnetic signal, for example, for the emission and absorption of
a radio signal. For these two events, 5 = in all reference systems,
because the velocity of light is invariant and we must always have
that l ct. Since the curve in Fig. 33 in actual fact corresponds to
3+1 dimensions (three spatial arid one time) instead of a plane, the
locus of zero intervals is picturesquely called "a light cone."
Proper time. The concept of proper time of a particle is closely
related to that of interval. This is the time measured in a coordinate
system fixed to a particle. The displacement of the particle relative
to this system is equal to zero by definition. Hence, the proper time
that has elapsed between any two positions of the particle is pro
portional to the interval calculated for these two positions:
(20.30)
Here, v is the velocity of the particle relative to an arbitrarily chosen
reference system in which the interval of time is equal to dt. From
(20.30) and also (20.21), the proper time is always the shortest.
For finite time intervals
t Q = ( dt 1/1  ~ < f dt, (20.31)
J t C J
i.e.,
t Q ^t. (20.32)
From (20.32) there follows a consequence with which, at first
sight, it is difficult to agree. Characteristic time is the time which
determines the rhythm of life processes in the human organism. And
for this reason if an imaginary traveller leaves the earth with a velocity
close to that of light, and later returns to the earth, then, in accordance
with (20.32), he will have grown less old than a person, initially
of the same age, remaining on the earth.
The asymmetry between the traveller and person that remained
on the earth is explained by the fact that the traveller was not moving
inertially he first travelled away and then returned. For this, it
was necessary for him to turn about in some way, i.e., to lose the
property of inertiality retained by observers on the earth. It may be
noted that the time spent in turning may make up an infinitesimally
small amount of the time of travel, if the journey itself is sufficiently
long. This is why the turning operation cannot in any way reestablish
equality between t and . But this operation is necessary in order
that the comparison between the ages of both observers can be
performed, i.e., to return them to the same point of space and to
a single coordinate system. Thus, disregarding the "striking" formu
lation of the traveller experiment, we may say that time in a non
206 ELECTRODYNAMICS [Part II
inertial system may differ to any extent from the time in an inertial
system, even though the noninertial system may deviate from
inertiality for an extremely short time.
The case of the human traveller is, of course, purely imaginary,
technically speaking. But a relationship of the type (20.32) is observed
in the decay of mesons in cosmic rays. The mean life of a positive
Tcmeson, of mass 273 electronic masses, in decaying to a jAmeson,
of mass 207 electronic masses, together with a neutral particle, is
2 x 10~ 8 sec (the negative Ttmeson is most often captured by nuclei).
This time is measured for a 7umeson stopped in the substance, i.e.,
it is proper time. The velocity of the meson, like that of any other
particle, does not exceed c. If the relationship (20.32), expressing
the relativity of time, did not exist, then a rapid remeson with a
velocity of the order of c would on the average travel through
c x 2 x 10~ 8 cm = 600 cm of air. Actually, the mean path of a 7umeson
is considerably greater due to the fact that its lifetime, in a co
ordinate system fixed in the air, is considerably greater than its
proper lifetime.
Frequency and wavevector transformations of electromagnetic
waves. Proceeding from invariants, we may find out in what way
the noninvariant quantities involved are transformed in passing
from one reference system to another. We shall now show that the
transformation properties of wavevector components and frequency
are those of coordinates and time.
In order to prove this, it is sufficient to note that the phase of a
wave is invariant. Indeed, the phase characterizes some event, for
example, the fading of the electric and magnetic fields at a certain
instant of time and at a certain point in space. If we examine this
wave in another coordinate system, then the coordinates and time
(corresponding to this event) will have other values, though the event
itself cannot change, of course. This is easy to understand if we
imagine that the electric and magnetic fields are measured by the
readings of some inertialess device. Two such devices, situated at
the same point of space at a certain instant, but in motion relative
to each other, must together indicate the zero value of the field.
Otherwise, the coordinate system in which the electromagnetic field
is equal to zero will in some way be distinguished from the
others.
From (17.21) and (17.22), the expression for the phase of a wave is
fy = xk x + yk y + zJc z <*t .
This quantity must be invariant in transforming to another coordinate
system. Let us express x' and t' in accordance with (20.17) and (20.18)
and substitute them into the condition of invariance of phase :
Sec. 20] THE THBOBY OF RELATIVITY 207
xk x + y ky + zkz u>t = x' k' x + y' k r ' + z' k^ co' t' =
Now comparing the coefficients of x, y, z, and t, we obtain the
transformation equations for the wavevector components and
frequency:
'
(20  33)
(20.34)
(20.35)
(20.36)
These are entirely analogous to equations (20.19) and (20.20).
The longitudinal and transverse Doppler effect. If the frequency
of a source of light with respect to its own coordinate system is equal
to co', and the angle between its velocity and the line of sight is
equal to &', so that k' x = lc' cos &' = cos <9', then we obtain from
(20.36)
c/(l H cos^'J
co = \ r _L Tg5 . 7  . (20.37)
In particular, if the source moves along the line of sight (i.e., towards
the observer),
(20.38)
F 2
c 2
These equations describe the wellknown Doppler effect, by means
of which the radial velocities of stars are measured. The square root
in the denominator gives a correction introduced by the theory of
relativity into the formula usually used.
If the velocity of the source is perpendicular to the ray, then,
from the requirement that k x = 0, we also obtain a change in frequency,
although it is of second order with respect to :
c
208 ELECTRODYNAMICS [Part II
(20.39)
This transverse effect was observed by Ives in the radiation from
moving ions (in canal rays) when the ratio was sufficient to detect
the frequency displacement spectroscopically. This gives direct
experimental proof of the contraction of the time scale in relative
motion.
A comparison of inertial forces and the force of gravitation. Let
us now investigate the transformation from inertial to noninertial
systems. We define the latter as a system in which there are inertial
forces.
All inertial forces have the common property that they are pro
portional to the mass of the body. Among the interaction forces,
only one force is known Avhich possesses that property this is the
force of Newtonian gravitation. The fact that gravitational force
is proportional to the mass of the body is very well known, though
very surprising all the same. The mass of a body may be defined
from Newton's Second Law when any kind of force (electric, magnetic,
elastic, etc.) acts on the body. It is therefore very difficult to under
stand why the force of interaction between bodies, namely, the
force of gravitation, is proportional to that very mass involved in
the expression for Newton's law [see (2.1)]. As is well known, all
other interaction forces are independent of mass.
In addition, the very form of the gravitation law itself somewhat
contradicts our physical intuition in accordance with this law,
gravitational forces are transmitted over any distance instantaneously.
Einstein called attention to the profound significance of the analogy
between inertial forces and the force of gravitation. In certain cases
these forces are indistinguishable in their action. For example, when
an aeroplane performs a turn and, in doing so, inclines the plane
of its wings, the passengers feel, as before, that the direction of gravity
acting on them is perpendicular to the floor of the cabin. In this case,
a resultant force consisting of gravity and a centrifugal force operates
like the force of gravity they are both proportional to the mass
and act on all the bodies inside the aircraft in the same way. It is
physically impossible to separate these two forces without con
sidering objects outside the aircraft.
When a lift begins to rise the gravitational force is, as it were,
increased the force of inertia due to the acceleration of the lift
is added to it.
In these examples (almost trivial), the inertial and gravitational
forces are equivalent "on a small scale," that is, in certain small
regions of space. In large regions, there is a certain essential difference
SeC. 20] THE THEORY OF RELATIVITY 209
between the behaviour of inertial forces and gravity. The latter
diminishes with the distance from the centre of attraction, while
inertial forces either remain constant or increase without limit. Thus,
centrifugal force increases in proportion to the distance from the
axis of rotation. The force of inertia in a coordinate system fixed
in an accelerating lift is the same at any distance away from the lift.
The general theory o! relativity. The basic idea of Einstein's gravi
tational theory is that motion in a gravitational field is the same
sort of inertia! motion as the accelerating of passengers relative
to a braking carriage. It is precisely for this reason that acceleration
due to the action of gravity does not depend on the mass of the body.
In order to understand why the force of gravitation, as opposed
to the wellknown inertial forces, becomes zero at an infinite distance
away from attracting bodies, we must assume that the space close
to the attracting bodies does not have the geometrical properties
of Euclidean space. In other words, we must take it that space and
time obey nonEuclidean geometrical laws in the sense of the ideas
first developed by Lobachevsky and later by Riemann. Free motion
in such non Euclidean (Riemamiiaii) space is curvilinear. However,
since it is precisely the i>roperties of space itself that determine
the curvature, acceleration of bodies does not depend upon their
masses (if we can neglect the effect of the latter on the gravitational
field) in the same sense as the field of a falling stone does not affect
the gravitational field of the earth.
Thus, gravitational and inertial forces are indistinguishable in
small regions of space. In such regions, a noniiiertial coordinate
system is equivalent to an inertial system in which an additional
gravitational field is operative, with the same acceleration of falling
bodies which, in the noninertial system, is ascribed to inertial forces.
For this reason, this theory of gravitation is also called the general
theory of relativity in contrast to the special theory of relativity,
which considers only inertial systems.
Since the equations of motion in the genera] theory of relativity
(in the same way as all equations of motion) are formulated in differ
ential form, equivalence on a small scale is quite sufficient for writing
down the equations.
However, we must remember that a rotating coordinate system
is not, as a whole, equivalent to a gravitational field. Indeed, a rotating
system can, generally, only be determined for distances from the
axis of rotation for which the velocity of rotation is less than that
of light. This is why a rotating system is not equivalent to a noii
rotating system, which has meaning in infinite space, too.
The mechanics of Einstein's general theory of relativity is con
siderably more complicated than Newtonian mechanics, 'which is
included in this theory as a limiting case. But Einstein's theory is
free from the gnosiological concepts, so alien to us, of the hypothesis
14  0060
210 ELECTRODYNAMICS [Part II
of action at a distance. The properties of space and time in Einstein's
theory are studied in inseparable unity with the motion of matter,
and not only as a requisite for the motion of matter. Abstract space
and time, which in Newtonian physics were sometimes regarded as
almost belonging to logical, a priori categories, do not exist in Ein
stein's gravitational theory in the general theory of relativity, space
and time are endowed with physical properties.
The consequences of Einstein's gravitational theory. Einstein's
refined gravitational theory leads to a series of results that may be
verified by astronomical observation.
1) The perihelion of Mercury should rotate through 43 * per century.
This is in excellent agreement with astronomical facts.
2) Rays of light from stars passing near the limb of the sun should
be displaced towards it, since light is not propagated rectilinearly
in nonEuclidean space. This result also agrees closely with accurate
observations made during solar eclipses.
3) Spectral lines in heavy stars should be shifted towards the red
end of the spectrum, and this, too, is found to be the case.
For the first time in the history of science, the general theory
of relativity made it possible to pose the cosmological problem, i.e.,
the problem of the structure and development of the Universe.
The present state of the cosmological problem is far from a solution
due to insufficient astronomical data and to the mathematical diffi
culties associated with Einstein's gravitational equations. It should
be noted that before the general theory of relativity, the cosmological
problem was posed in a purely speculative way; Einstein's theory
indicated the path for scientific investigation and has led to a series
of important results.
Exercises
1) Calculate the change in the velocity of light propagated through flowing
water in Fizoau's experiment.
Disregarding the theory of relativity, the result would be u J =  V .
2) Obtain a precise equation for the aberration of light, with an arbitrary
inclination & of the ray of the ecliptic.
cos 0
Answer: cos&'^ = .
1 cos &
c
3) Write down the equations for the Lorentz transformations for an arbitrary
direction of the velocity V relative to a coordinate system.
Sec. 21] RELATIVISTIC DYNAMICS 211
rV r'V
In our equations x  ^ , .c' = . The component perpendicular to the
velocity is
, V(rv) , VJr'V)
F 2  r y*
From (20.17)
21V,
^V _F
V ~ "I/ 72" '
[/ ^2"
y
Multiplying this equation by == and adding equations, we obtain
4) Write down the Lorentz transformation equations for the components
of acceleration.
5) Show that the "fourdimensional volume element" dx dy dz dt is invariant
with respect to a Lorontz transformation.
(f/2 \ i/ / 1/2 \ i/a
1  ^1 from (20.22) and dt = dt' 1 1  ~\
from (20.21), whence tho statement follows.
(>) A light beam is within a solidangle element rfQ. Show that Lorentz
transformations leave the quantity co 2 ^Q invariant.
Uso the result of exercise 2: dQ. = 2nd cos ft.
Sec. 21. Relativistic Dynamics
Action for a particle in the theory of relativity. The adjective rel
ativistic denotes invariaiiee with respect to Lorentz transformations,
which invariance satisfies the relativity principle. For example,
Maxwell's equations in free space are relativistic.
In effect, the Lorentz transformation is derived from the require
ment that the equations of electrodynamics remain invariant. There
fore, the proof of the relativistic invariance of Maxwell's equations,
which proof will be given somewhat later in this section, is simply
in the nature of a confirmation.
The situation with mechanics is altogether different. Newtonian
mechanics satisfies only the Galilean relativity principle, which holds
for velocities small compared with c. Therefore, it is necessary to
find equations of mechanics such that they will be invariant with
respect to Lorentz transformations.
In Sec. 10 it was shown how to develop a mechanics by proceeding
from the principle of least action. And it was found possible to deter
mine the form of the Lagrangian of a free particle by proceeding
from two basic assumptions [see equations (10.11)(10.13)]:
14*
212 ELECTRODYNAMICS [Part II
1) Action is invariant to Galilean transformations;
2) The Lagrangian of a free particle depends only on the absolute
value of velocity; the velocity vector v cannot be involved in it
because, in the absence of an external field, there are no distinguishable
directions (in space) relative to which the vector v can be given.
In relativistic mechanics the first condition is replaced by the
in variance to a Lorentz transformation, while the second condition
remains unchanged. Both conditions are satisfied by an action function
of the form
8 = J a ds = j ac /1^ dt , (21.1)
where we have used the relationship (20.30) between ds and dt. Agree
ment with the first condition can be seen from the fact that action
is expressed in terms of interval only, while agreement with the
second condition is obvious. No other invariant quantities can be
constructed from dl and dt except the interval, whence the uniqueness
of the choice (21.1).
The Lagrangian for a free particle. In order to define the constant a,
we examine the limiting form of (21.1) for a small particle velocity.
If v<^c,
From the definition of the Lagrangian (10.2)
S = JLdt (21.3)
it follows that the Lagrangian is
____ 2
(21.4)
The first term in (21.4) is constant and can be omitted as not appearing
in Lagrange's equation [see (10.8)]. The second term should be
compared with the Lagrangian for a free particle in Newtonian
mechanics :
L=2f. (21.5)
Whence
K^ mc. (21.6)
The meaning of m here is the mass of the particle measured in a
coordinate system in which the particle is at rest (or infinitely near
rest). Thus, by its very definition, the quantity m is relativistically
invariant. Finally, we have the Lagrangian in the form
 . (21.7)
Sec. 21] RELATIVISTIC DYNAMICS 213
Momentum in relativistic mechanics. From (21.17), we immediately
obtain an expression for momentum in the theory of relativity:
(21.8)
As required, at small particle velocities it reduces to the Newtonian
expression p = rav.
Sometimes the quantity (i.p.. ? the proportionality factor
V 1 2 /c 2
between velocity and momentum) is called the mass of motion of
the particle, as opposed to the rest mass m. To avoid confusion we
will not use the expression "mass of motion," and will take the
term mass to mean the quantity m which is relativistically invariant
by definition.
The limiting nature of the velocity of light. The limiting character
of the velocity of light, about which we have already spoken in
Sec. 20, can be seen from equation (21.8). As the velocity of a particle
approaches the velocity of light, its momentum tends to infinity.
The only exception is a particle whose mass is equal to zero. Its
momentum, written in the form (21.8), gives the indeterminate form
0/0 for v c and can remain finite. But then the velocity of this
particle must always equal c. This property, as we know, is relativist
ically invariant since the velocity of light is the same in all inertial
systems. The momentum of such a particle must be given in
dependently of its velocity [and not according to equation (21.8)],
since the velocity is already determined and is equal to c. A velocity
greater than c is utterly meaningless because it involves an imaginary
quantity for momentum.
Energy in the theory of relativity. Let us now determine the energy
of a particle. In accordance with the general definition for energy (4.4),
T
Z,=
ys
Equation (21.9) once again confirms the limiting nature of the veloc
ity of light. When v tends to c, the energy of the particle <f tends to
infinity. In other words, an infinitely large quantity of work must be
performed in order to impart to the particle a velocity equal to that
of light.
Best energy. From equation (21.9), the energy of a particle at rest
is equal to me 2 . Let us apply this equation to a complex particle ca
pable of spontaneously decaying into two or three particles. Many
atomic nuclei and also unstable particles (mesons) are capable of such
disintegration. In the disintegration, the energy must be conserved,
214 ELECTRODYNAMICS [Part II
< ? = <? 1 + < f 2 , (21.10)
because disintegration is spontaneous, caused not by any external
interaction, but by some internal motion in the complex particle.
The Lagrangiaii for this motion is not known explicitly, but in any
case it cannot involve time. Therefore, the energy of a complex particle
before disintegration is equal to the energy of the two particles formed
after the disintegration, when there is no longer any interaction
between them.
The energy of all these particles is expressed in accordance with
equation (21.9), as applied to all free particles (whether simple or
complex) when their motion is considered as a whole. The only possible
form of the Lagrangian for such motion is (21.7), from which it follows
that the energy is in the form (21.9). Substituting this expression in
(21.10) and noting that the initial particle was at rest, we obtain
.. (21.11)
But the terms S^ and <^ 2 on the right are correspondingly greater than
W]C 2 and ra 2 c 2 , whence we obtain the fundamental inequality
m^m 1 + m 2 . (21.12)
Hence the mass of a complex particle capable of spontaneous dis
integration is greater than the sum of the masses of its component
particles. In Newtonian mechanics, the mass characterizing the motion
of the system as a whole [see the last term of equation (4.17)] is equal
to the sum of the masses of the component particles.
If we define the difference
_ mc * (21.13)
as the kinetic energy of a particle (for small energies it reduces to
T = ^Tr 1 and call me 2 the rest energy, then it can be seen from the law
of conservation of energy (21.11) that part of the rest energy of a
complex particle is converted into kinetic energy of the component
particles, and part is converted into their rest energy. Only the total
energies 6\ and not the kinetic energies 2 7 , satisfy the conservation
law because the kinetic energy of a complex particle as a whole is
equal to zero before disintegration and cannot be equal to the essen
tially positive kinetic energy of the disintegration products.
In chemical reactions, the change in the rest masses of the reacting
substances occurs in the order of 10~ 9 (and less) of the total mass.
SeC. 21] BBLATIVISTIC DYNAMICS 215
In nuclear reactions, where the particle velocities are of the order c/10,
the change in mass may approach one per cent.
When an electron and positron (a positive electron) are annihilated,
their energy, including rest energy, is totally converted into the energy
of electromagnetic radiation.
As we shall see from quantum theory, radiation is propagated in
space in the form of separate particles socalled light quanta (quan
tum mechanics teaches that this is compatible with the wave proper
ties of radiation!). The velocity of a light quantum is equal to c so
that its mass is identically equal to zero. For this reason, the total
rest mass of the particles taking part in the annihilation process is
2 me 2 before the annihilation and zero afterwards.
However, the change in the energy of the electromagnetic field is,
of course, equal to 2 me 2 , provided the electron and positron did not
have any additional kinetic energy. We could, by convention, call the
energy of an electromagnetic field, divided by c 2 , its mass. With such
a definition of mass, the total "mass" would be conserved. But com
pared with the law of conservation of energy, such a law of conserva
tion of "mass" does not contain anything new; it only repeats the law
of conservation of energy in other units.
It is precisely the rest mass that is best to use in describing nuclear
reactions, for a change in rest mass determines the energy which may
be generated as a result of the reaction (in the form of kinetic energy
of the disintegration products, or in the form of radiated energy).
There is no sense in calling the energy of a light quantum divided by
the square of the velocity of light, its mass, because this quantity
does not in any way characterize light quanta. Tin's quantity has one
value in one reference frame and another value in another frame,
because the energy of any particle depends upon the reference system
relative to which its motion is defined. Yet rest mass is a quantity
that characterizes the particle. For example, the resfc mass of an elec
tron, involved in the expressions for all its mechanical integrals of
motion, is equal to 9 x 10~ 28 gm. The corresponding quantity for a
quantum is identically equal to zero, and, in this sense, characterizes
a light quantum in the same way that the quantity 9 x 10~ 28 gm is
characteristic of an electron.
The mass of a particle determines the relationship between the
momenttim and velocity of the particle in accordance with equation
(21.8). It is impossible to determine the mass of a particle by its
momentum alone, since particles with the same momenta can have
quite different masses. For this reason, it is meaningless to state
(though this is sometimes done) that the existence of light pressure
(i.e., momentum of the electromagnetic field) proves that the light
quantum has a finite mass.
It is sometimes said that a mass of one gramme is capable of releasing
an energy of 9 x 10 20 ergs (i.e., 1 c 2 ). However, if the substance con
216 ELECTRODYNAMICS [Part II
sists of atoms the possibility of generating this energy is still question
able since up to now not a single process is known in which the total
quantity of protons and neutrons (collectively called, nucleons) is
changed.* This is why, the relative change in rest mass in nuclear
reactions is always measured in fractions of one percent.
The possibilities of various reactions are also limited by the conser
vation of total charge.
The Hamiltonian for a free particle. We shall now express energy in
terms of momentum. Squaring equation (21.9) and subtracting from it
equation (21.8), after it has been squared and multiplied by c 2 , we
obtain
^2_ c 2p2 ==m 2 c 4  (21.14)
We have called the energy expressed in terms of momentum the
Hamiltonian [see (10.15)]. Hence,
p* . (21.15)
Whence we obtain a relationship between the energy and momentum
of a particle that has no rest mass :
= cp. (21.16)
The Lorentz transformation for momentum and energy. We shall
now find out how energy and momentum behave with respect to a
Lorentz transformation. From equation (21.8) we get
mv x mdx dx
ds '
dy dz
ds ' " z ds '
mc 2 dt o dt
"~ r=mc "57
(21.17)
The quantities m, c and ds are invariant. Hence, the components
p x , p y and p z are transformed similar to dx, dy, and dz, i.e., similar
to x, y, and z. In accordance with the last equation, energy transforms
like time. We can make the following comparison: p x ~ x, p v ~ y y
Pz~ z, d? ~c 2 t.
* In order to annihilate the whole mass we would have to first prepare
"antimatter" (when ordinary matter interacts with antimatter they are mutually
annihilated, cf. Sec 38). But this would require a like expenditure of energy.
SeC. 21] BBLATIVISTIC DYNAMICS 217
Now substituting momentum and energy in the Lorentz transfor
mation (20.17) and (20.18), we obtain
(21.18)
Pi = P* , (21.20)
We note that a correct transition from (21.18) to a nonrelativistic
equation for the transformation of energy is obtained only when the
rest energy me 2 is substituted in place of $' , for then p' x = p x mV
(i.e., v' x v x F) in agreement with the Galilean law for addition of
velocities.
Hence, if we demand that the Lorentz transformation yield the
correct limiting transition to a Galilean transformation, it is necessary
to include the rest energy of the particles in their total energy. Con
versely, the kinetic energy T (21.13) does not give a correct limiting
transition.
Further, we note that if we form the expression <? 2 c 2 p 2 from equa
tions (21.17) we obtain
c p m c ^ jji j me
in accordance with (21.14)
The velocity of a system of particles in the theory of relativity. We
shall now show how to determine the velocity of a system of particles
in relativity theory. We shall consider two particles. Between the veloc
ity, momentum, and energy of each particle there exists the relation
p^. (21.22)
It is obtained if we divide (21.8) by (21.9). The same equation can
also be obtained somewhat differently. Let us determine, from (21.18),
the velocity V of the coordinate system relative to which the momen
tum of the particle is equal to zero. Putting p' x = on the lefthand side
of (21.18) we will have, on the right,
or, if the velocity is not along the #axis at all, in accordance with
(21.22) V=pc 2 /<? = v. As applied to a single particle, the statement
218 ELECTRODYNAMICS [Part II
v== V is trivial and simply denotes that the momentum of the particle,
relative to a coordinate system moving with the same velocity as the
particle itself, is equal to zero.
We now apply equation (21.18) to two particles in order to find the
velocity of the coordinate system relative to which their total momen
tum is equal to zero. The total momentum in the primed system is
pi+pip', and the total energy tf\ +<?' 2 <?' Let us take the zaxis
along p'. Since tho Lorentz transformation is linear and homogeneous,
it has the same form for the sum of two quantities as for each separate
ly. Therefore, we immediately obtain an equation similar to (21.22):
v __ ''(Pi ! J>i) /<>! 23)
*  *r\ *r m ( }
Tho primes may be omitted here. In order to obtain the limiting tran
sition to the velocity of the centre of mass in Newtonian mechanics
from (21.23), it is necessary to take Pi M^, P 2 = w 2 v 2 , ?i = w t c 2 ,
^ 2 = m 2 c2 > i e > ^ no Particle energies are replaced by their rest energies.
Tho quantity V, expressed in terms of the particle velocities accord
ing to equations (21.8) and (21.9), does not have the form of a total
derivative of any quantity with respect to time. Therefore, it is im
possible in relativistic mechanics to determine its coordinates in terms
of the velocity of the centre of mass. It is better to say that if we
attempt to express the coordinates of the centre of mass by means of
a classical (or some other) equation it is impossible to represent V
in the form of a time derivative of these coordinates, except in the
trivial case when \ l and v 2 are constant. This is why the concept of
centre of mass for particles moving in accelerated motion cannot be
used.
As regards relative velocity, v x  V 2 , it is meaningless in. relativistic
mechanics, since there is no simple law for the combination of veloc
ities.
Action for particles in an electromagnetic field. Let us now turn
to the equations of motion for a charged particle in an electromagnetic
field. We already know the part of the action function which describes
the interaction of charges and Held. From equation (13.17), this is S v
Since the variation of S l leads to Maxwell's equations, we can be cer
tain of the relativistic invariaiice of /S^. As applied to point charges,
we have already written S 1 in magnctostatics in equation (15.26).
But action for free charges in the next equation, (15.27), was suitable
only for small particle velocities.
We now know the Lagrangian for a fast particle in the absence of a
field (21.7). Thus, the Lagrangian in an external field is equal to the
sum of the relativistically invariant expressions (21.7) and (15.26):
P= wic 2 l ~ +~Ave<p. (21.24)
Sec. 21] BELATIVISTIO DYNAMICS 219
We now obtain an expression for momentum and energy. Momen
tum is
(21.25)
Here, p denotes momentum in the absence of a field.
From (4.6), the energy is
(21.26)
where <O Q is the energy in the absence of an external field; according
to (21.9), it is equal to
Thus, the linear term in velocity does not appear in the energy
expressed in terms of momentum. It will be seen here that the Lagran
gian is not of the form T U because it involves the linear term  A v.
The Hamiltonian for a charge in an external field. From (21.25)
we obtain
Po = plA, (21.27)
and, from (21.26),
^ ='C9. (21.28)
But we already know the expression for S J Q in terms of p from equa
tion (21.15), which relates to the energy and momentum of a free
particle. Substituting p and S\ in (21.15) in accordance with the
last equations, we obtain the Hamiltonian of a charge in a field:
Jf = y m 2 C 4 + c 2 (p   A)'  e 9 . (21.29)
The equations of motion of a charge in an external field. From (21.29)
we can obtain the equations of motion for a charge in an external
field. However, it is simpler to make use of the Lagraiigiaii (21.24).
We know that Lagrange's equations are of the following form:
d *L dL
~dt~d7 ~aF U ' ( 21  3 )
where equation (21.30) replaces three equations of the form (2.21)
for the coordinates of v and r.
220 ELECTRODYNAMICS [Part II
The derivative ~ is equal to p^p +~A, so that its total time
G V C
derivative is
d dL dp , e dA /^i o^\
dt dv dt c dt '
In order to expand the expression = , we first write it down for
one component:
dA x __ dA x dA x dx^ , dA y dy_ dA z dz^ __ dA x , _. ^
dt ~~ dt "" dx dt ~*~ dy dt ~^~ dz dt ~~ dt ^ ' *
(21.32)
[see (11.31)], whence, going to a vector equation and substituting
in (21.31), we have
^1^ == ^ + A/^L + (vV) A) . (21.33)
Let us now calculate the righthand side of (21.30). Instead of ^ we
can write the completely equivalent expression VL:
The gradient V (Av) denotes coordinate differentiation, where only
A and not v depends explicitly on the coordinates. Thus, applying
o r
equation (11.32), we find ~ :
~ s Vi = (W) A +  [vrot A]  e<p . (21.34)
o r c c
Now, substituting (21.33) and (21.34) in (21.30) and taking all the
terms involving potentials to the righthand side, we obtain
(21.35)
The righthand side of (21.35) involves the electromagnetic fields in
accordance with their definition in terms of potentials (12.28) and
(12.20).
Hence, the equation of motion of a charge involves only the field
and not the potential, as follows from the condition of gauge invariance.
After substituting the fields the equation takes the form
d mv ., , e r TT T
+[
The righthand side of (21.36) is called the Lorentz force. In addition
to the usual term, eE, which we know from electrostatics, it involves
Sec. 21] BELATIVISTIC DYNAMICS 221
a term similar to the Coriolis force. It is related to the part of the
Lagrangian which is linear in velocity.
The magnetic part of the Lorontz force, [vH], is very similar to the
expression for the force acting on a current in an. external magnetic
field and, naturally, can be obtained from it. We did not have to use
this method of derivation because the part of the Lagrangian which
describes the interaction between charges and field was already known
from Sec. 13. And besides, the relativistic invariance of (21.36), which
emerges obviously from derivation from the invariant Lagrangian
function, is considerably more difficult to grasp from the elementary
definition of a magnetic force acting on a current.
The work performed by a field on a charge. From equation (21.36),
we can obtain an expression for the work done by an electromagnetic
field on a charge. We know by definition that the work is equal
to the change in kinetic energy. Let us multiply scalarly both parts
of (21.36) by v. We shall then have the expression ~ on the left
hand side. But v^  inaccordancewithHamilton > sequation(10.18),
so that v ^5 = ? 4~ = f  = ~Jr > and on the lefthand side what
dt dp dt at dt
Ave have is the required quantity for the change of kinetic energy
in unit time. On the righthand side the term v [v H] = [vv] H =
and there remains only the work done by the electric force :
As was to be expected, the magnetic force [vH] does not per
form work on the charge because it is perpendicular to the charge
velocity at every given instant of time.
The Lorentz transformation for the field components. From (21.37)
and equation (21.36), if we write it in terms of components, it is
easy to obtain the Lorentz transformation equations for the field
components. These equations must be written so that their form
does not change in passing from one coordinate system to another.
Let us take equation (21.36) for the component of momentum on
.r, and multiply it by~. We shall also multiply equation (21.37)
fj t v
by 7 and also by ^ , where V is the relative velocity of the coordinate
system. After this we subtract (21.37) from (21.36). Then on the left
hand side we have
222 ELECTRODYNAMICS [Part II
On the righthand side we will have the expression
/>/F dt _L l IT d V l JJ dz \ eV Iff dx _l_ W dy _L F dz \
e \rj x ^ HZ 5 kl v~j~l 2 \" J xj r J^y ~3 r J^z ~r~l ~
\ ds c ds c y ds J c 2 \ da y ds ds J
oW tdt V dx\ el n V \<ly <>/ L Vp\dz
= & rjx\r 9 r~"l i \"x " J vt~~f l^*v~^ ^*I^T~
\ds c 2 ds I ' c\ c y / ds c\ y ' c / ds
But ds is invariant. Therefore, the quantities on the righthand
side must be transformed in accordance with the basic equations
(20.17) and (20.18). Differentiating these equations, we obtain
dt __ V rZ.r __ rZ,
ds c 2 ds ds I/ c 2 ' ds ds * ds ds
Now, dividing both sides of the equation by /I r and multi
d s I
plying by ^ r , we will have an equation for the ^component of
momentum in the new coordinate system:
In accordance with the principle of relativity, this equation must
be written in the same way as for the unprimed coordinate system:
*?'* _,,#/. e ji< <W G pr> dz '
dt' "" *' jj * ^ i 11 * 'dt' ~~ c u v dV '
Comparing the last two equations, we obtain the field transformation
equations:
E' X =E X , (21.38)
77 y + JL^
H} __r ., (21.39)
V 1 ^
  Ey
(21.40)
In the same way, though from other equations (21.36), is it easy
to find other equations for field transformation:
H X =H X , (21.41)
Sec. 21]
RELAT1VISTIO DYNAMICS
223
~\
(21.43)
Consequently, in contrast to coordinates, it is not the longitudinal
but the transverse components that are transformed in the field.
The change in field, in passing from one
coordinate system to another, is verified to a
nonrclativistic approximation (i.e., to the ac
pr\ \
curacy of terms of the order \ in a unipolar
induction experiment. A diagram of the ex
periment is shown in Fig. 34. The magnet NS
rotates around its longitudinal axis. Two col
lectors connected by a fixed conductor are joined
to the centre of the magnet and to its axis.
When the magnet is rotated, an e.m.f. appears
in the wire. This experiment is frequently inter
preted as meaning that wlieu the magnet is
rotated the wire "cuts" its lines of force as
if the lines were attached to the magnet like
brushes.
Actually, unipolar induction must be understood as follows. There
is only a magnetic field H in the coordinate system attached to the
magnet, while the electric field is equal to zero. Heuce, in a system
fixed in the wire, relative to which the magnet moves, an electric
field, too, should be observed in accordance with (21.42) or (21.43).
This field is of an order of magnitude  // and produces the e.m.f.
We note that a coordinate system is defined only when both the
electric and magnetic fields are specified. It is insufficient to specifiy
only one of them.
The invariants of an electromagnetic field. From equations (21.38)
(21.43), it is easy to obtain the following two invariants:
Fig. 34
E'*  #' 2  E*  H 2 ,
E'H'EH.
(21.44)
(21.45)
From the invariance of these expressions it follows that the electro
magnetic field of a plane wave appears similar in all systems. Indeed,
in a plane wave, E = H or E 2 H 2 = Q. This property is invariant
according to (21.44). Further, E j_ II, so that (E H ) 0. This
property is invariant according to (21.45).
The quantity (E H) is invariant with respect to a Lorentz trans
formation. But with respect to a replacement of x, y, z by #, y, z
224 ELECTRODYNAMICS [Part II
(i.e., an inversion of the signs of the coordinates), it is not invariant,
because in this case E changes sign while H does not (see Sec. 16).
The quantity E 2 H 2 is invariant even when the coordinate signs
are inverted. But this quantity is the Lagrangiaii for a free electro
magnetic field. Integrated over the invariant volume dx, dy, dz, dt
(exercise 5, Sec. 20), it yields invariant action, as required, while
the quantity (EH) does not give a real invariant.
The linearity of Maxwell's equations with respect to field. A real
invariant can be formed from the quantity EH merely by squaring.
It is, of course, not at all obvious beforehand why such a quantity
as well as the square of the invariant E 2 H 2 cannot appear in the
Lagrangiaii for an electromagnetic field. The same can be said of
higherorder terms which do not change sign in the substitution of
x by x, etc. But if some terms other than quadratic with respect
to field are left in the Lagrangiaii, then Maxwell's equations will
contain nonlinear terms.
The essential difference between nonlinear and linear equations
is that the sum of two solutions of a nonlinear equation is not its
solution. Indeed, if two electromagnetic waves are propagated in
a vacuum they are simply combined, and in no way distort each
other. In nonlinear theory, the velocity is a function of the wave
amplitude, while in electrodynamics the velocity of light is a universal
constant.
For this reason, the choice of the Lagrangian in the simplest form
E 2 H 2 expresses the experimental fact that the law of variation
of any electromagnetic field in space and in time is in no way dependent
upon whether another field is operative in that same chargefree
region of space.
Actually the quantum electromagnetic field theory indicates the
existence of certain nonlinear effects. In the range of phenomena
for which classical electrodynamics is applicable, these effects are
not essential.
Transformation of charge density and current density. From the
definition of charge density one can find the law of its transformation.
Since charge is an invariant quantity, we have
de = pdxdydz = Pod# d?/ dz , (21.46)
where p is the charge density in a system relative to which it
is at rest and, hence, the quantity is also invariant by definition.
Whence,
dx dy dz dt dt /rtl ..
P = Po = Po d = Po c 37 ' ( 2L47 >
where we have used (20.30) and exercise 5, Sec. 20. The current
density is
SeC. 21] BELATIVISTIC DYNAMICS 225
dx dt dx
j, = ptfc = Po c gj  dT = Po 37 ,
From here it can be seen that the current components are trans
formed like coordinates, while charge density is transformed like
time.
Let us consider a conductor in which a current is flowing. In the
coordinate system in which the conductor is at rest, it remains neutral,
but in other systems a charge density must appear on it. This fact
does not contradict the invariance of total charge, but follows from
it in accordance with (21.46)(21.48).
The invariance of action for the field. Let us now verify that the
action term (13.17) describing the interaction of field and charge
is invariant. It follows from (21.27) and (21.28) that the vector
potential transforms like momentum (i.e., like a radius vector),
while the scalar potential transforms like energy (i.e., like time).
For this reason the product
Aj 99
behaves, with respect to a Loreiitz transformation, like an interval;
in other words, it remains invariant. Integrated over the invariant
"fourdimensional" volume dx dy dz dt, it yields the invariant action
term $ x . Hence, Maxwell's equations are obtained from the invariant
action function S, so that they are also invariant themselves. This
could also have been verified from equations (21.38)(21.43).
Exercises
1) Find the scalar and vector potentials of a freely moving charge.
In its own coordinate system, the scalar potential is 9 = and the vector
r o
potential is equal to zero. Hence, in a system relative to which the charge
moves, its scalar potential is
9o e
9 = ' "
and the vector potential is
v
A =
15  0060
226 ELECTBODYNAMICS [Part II
Further, r must be expressed in terms of coordinates in the fixed system
(x  vt)
We can put instead of ttf, i.e., the abscissa of the moving charge. The
electromagnetic disturbance arrives at the given point x, y, z from the point ',
where the charge was situated earlier. We have
from the definition of lag. Putting 5 = vt in r , we obtain an expression for 9
and A in terms of JR'.
2) Find the motion of a charge in a constant uniform magnetic field.
If the field is in the direction of the 2axis, the equations of motion are
of the following form:
dp x _ e dy dp y _ __ e _^_jr dp z _
dt c dt ' dt c dt ' dt
Further, p 2 = const, p = const, pj + p} = const,
We look for the coordinates x and y in the form:
For R and co the following expressions result:
_ &v ecH
M "~^H 9 <0 ~~ r ~
The particle moves along a helix. For small velocities, to reduces to the
constant value .
me
3) Find the motion of a charge in a constant uniform electric field. The
equations of motion are
J** *JB *^Lc\ *V*H *#<Lw.*^d?L
From the last equation we obtain
*\/m 2 c* 4 c 2 (p ~f p$ + p) ^/m 2 c 4 f c a (pJ + pJ f p! )
From the first equation
These equation integrals together give x as a function of t.
Sec. 21] BELATIVISTIC DYNAMICS 227
If pz 9 = 0, then dividing p x by p y> wo have an expression for 5 in terms
of x (by eliminating t from the energy integral). The trajectory is of the form
of a catenary.
4) Find the motion of a charge in a central attractive Coulomb field.
The energy integral is of the form
Further, denoting the azimuth by ^, we obtain
dr m
  _ __ ff\ _
dt ir = ^r ' Pr ~ dt
FT'
Whence
1 dr __ p r
~~~
Substitution of p r in the energy integral and separation of the variables r
and fy leads bo an elementary quadrature. The trajectory for finite motion
(& < we 2 ) is similar to an ellipse, but with a rotating perihelion.
5) Examine the collision of a travelling particle of zero mass with a particle
of mass m at rest. Determine the energy of the incident particle after collision,
if its angle of deflection # is known.
Answer: $' '
H
6) Find the motion and radiation of a charge connected elastically to
some point of space (with frequency o> ) and situated in a uniform magnetic field
H Z = H, Hx = H y = 0.
The oscillations of the charge are governed by the following nonrelativistio
equations :
\  Hy t
c
 Hx,
c
mz= mcojjz.
The third equation does not depend on the first two. The first two equations
are easily solved if we put x = aeit*t, y = bei<*t f Then
a(cog6> 2 ) *co 6 = 0,
?nc
b (co <o 2 ) + iv>  a 0.
Let us multiply the second equation by i and first subtract it from the
first, and then add it. The combinations aib then satisfy the equations
15*
228 ELECTRODYNAMICS [Part II
Cancelling aib, we arrive at the equations for frequencies
e//co
'*^ = 0.
We regard  as small compared with co , replace co by co in the term  ,
and represent the difference to* to 2 as (to \ co ) (to co ), which is approximately
equal to 2oj (to to ). Then we obtain expressions for the frequencies of both
oscillations :
They differ from the undisplaced frequency by ' , i.o., by the Larmor
frequency COL.
From the equations for a and 6, after substituting co c^T COL, we have,
to the same degree of approximation,
a = i b .
If we represent the coordinates in real form, we obtain for both oscillations
x  a cos (to ^ COL) t ; y = a sin (to ^ COL) .
Thus, the radius vector of a particle performing oscillations with frequency
to 4 L rotates in a clockwise direction, while for oscillations with frequency
to a>L it rotates in an anticlockwise direction. Thus, in accordance with
Larmor's theorem, the frequency COL is added to the frequency co or subtracted
from it, depending upon the direction in which the charge rotates (wo note
that the sign of COL is changed for a negative charge).
Let us consider the radiation of such a charge in a magnetic field. We know
that the electric vector of the radiated electromagnetic wave lies in the same
plane as the charge displacement vector. Jf radiation is observed to bo due
to the zcomponent of the dipole moment, its electric vector is along the zaxis
and is proportional to 5 . Thus, the radiation is planepolarized and is of frequency
co . The oscillation occurring along the fiold and having an undisplaced frequency
radiates electromagnetic waves, which are polarized in the same plane as the
magnetic field. This oscillation does not radiate at all in the direction of the
magnetic field, but the oscillations with frequency co COL radiate circularly
polarized waves, and the electric field vector rotates in the same direction
as the charge displacement vector.
All three frequencies radiate in a direction perpendicular to the field. How
ever, since the charge oscillations are viewed from one side in this position
in circular rotation, the vector of electric field oscillations lies in a plane per
pendicular to the constant external magnetic field, so that waves with fre
quencies co t COL are now also planepolarized. In observations that are not
at right angles to the field, we obtain ellipticallypolarized oscillations and a
planepolarized oscillation of frequency c.> () .
The calculations set out here form the classical theory of the Zeeman effect.
The line splitting that is actually observed for various values of magnetic
field is correctly described only by quantum theory (Sec. 34).
PART III
QUANTUM MECHANICS
Sec. 22. The Inadequacy of Classical Mechanics.
The Analogy Between Mechanics and Geometrical Optics
The instability of the atom according to the classical view. Ruther
ford's experiments in 1910 established that the atom consists of
light negative electrons and a heavy positive nucleus of dimensions
very small compared to the atom itself (see Sec. 6). For such a system
to be stable, it is necessary that the electrons should revolve around
the nucleus like planets about the sun, for unlike charges at rest
would come together.
This stability condition of the atom is, nevertheless, insufficient.
In the case of motion in an orbit, electrons will experience centri
petal acceleration, but, as was shown in Sec. 19, a charged particle
undergoing acceleration radiates electromagnetic waves, thereby
transmitting its energy to the electromagnetic field. Thus, the energy
of an electron moving around a nucleus should continuously diminish
until the electron falls onto the nucleus. This statement is in striking
contradiction to the obvious fact of the stability of atoms.
The Bohr theory. In 1913, N. Bohr suggested a compromise as a
way out of this difficulty. According to Bohr, an atom has stable
orbits such that an electron moving in them does not radiate electro
magnetic waves. But in making a transition from an orbit of higher
energy to one with lower energy, an electron radiates ; the frequency
of this radiation is related to the difference between the energies
of the electron in these two orbits by the equation
h co = ^ <^ 2 >
where A is a universal constant equal to 1.054X 10~ 27 ergsec.
Both of Bohr's principles were in the nature of postulates. But
it was possible with their aid to explain, in excellent agreement with
.experiment, the observed spectrum of the hydrogen atom and also
the spectra of a series of atoms and ions similar to the hydrogen atom
230 QUANTUM MECHANICS [Part III
(for example, the positive helium ion, which consists of a nucleus
and one electron). Despite the fact that both of these, essentially
quantum, postulates of Bohr were completely alien to classical
physics and could in no way be explained on the basis of classical
concepts, they represented an extraordinary step forward in the
theory of the atom.
Indeed, the first postulate contains the statement that not every
state of the atom is stationary, but only certain states. This statement,
as we now know, derives from quantum mechanics just as directly
as elliptical planetary orbits derive from Newtonian mechanics.
The Bohr theory was very successful in explaining the spectra
of singleelectron atoms. But the very next step, a twoelectron
atom such as the helium atom, did not yield to consistent calculation
by the Bohr theory. The theory was even less capable of explaining
the stability of the hydrogen molecule. For this reason, the situation
in physics, notwithstanding a number of brilliant results of the Bohr
theory, was completely unsatisfactory. Besides the particular diffi
culties that we have noted here, the Bohr theory was, on the whole,
eclectic, since it was inconsistent in its combination of classical and
quantum concepts.
Light quanta. The inadequacy of classical ideas appeared most
obvious in the problem of the stability of the atom. But earlier
there were many facts which classical (i.e., nonquantum) physics
failed to explain. A case in point is the theory of an electromagnetic
field in equilibrium with matter (for more detail, see Sec. 42). Here,
classical theory leads to an absurd result the total energy of an
electromagnetic field in equilibrium with radiating matter is expressed
in the form of a divergent, i.e., infinite, integral.
In order to give a satisfactory description of experimental facts,
Planck, in 1900, postulated that sources of radiation emit and absorb
energy of the electromagnetic field in finite amounts. These discrete
quantities, or quanta, as they were called by Planck, are proportional
to the frequency of the emitted or absorbed radiation. It is easy to
see that the factor of proportionality must be the same as in Bohr's
second postulate (actually, Planck introduced a quantity 2 77 times
greater, but used a frequency v =  , equal to the number of oscil
lations per second). Bohr's second postulate relates the properties
of discreteness of stationary states of a radiating system (atom),
occurring in line spectra, to the energy of the emitted quanta. Classi
cally, it is just as impossible to explain this discreteness as it is to
explain Planck's initial hypothesis.
The duality of electrodynamical concepts. At the beginning of
the twentieth century, the classical theory of light also turned out
to be incapable of explaining many facts without appealing to an
additional hypothesis concerning light quanta. But at the same time,
Sec. 22] THE ANALOGY BETWEEN MECHANICS AND GEOMETRICAL OPTICS 231
there was a whole range of phenomena, such as diffraction and inter
ference of light, which appeared to be intimately bound up with
its wave nature. It did not seem possible to explain these phenomena
in terms of classical corpuscular concepts.
Another group of phenomena could be explained only on the basis
of Planck's hypothesis concerning light quanta, and was in most
obvious contradiction to the classical wave conceptions. Let us note
two such phenomena.
I refer, firstly, to the socalled photoemissive effect, i.e., the emission
of electrons from the surface of a metal in a vacuum when the metal
is illuminated by ultraviolet rays. The energy of the photoelectrons
depends only on the radiation frequency and is independent of its
intensity. This can only be understood if we assumed that the energy
of electromagnetic radiation is absorbed in the form of quanta, /tco.
Then the kinetic energy of an electron will be equal to the energy
of the quantum minus the electronic work function (the energy
needed to remove the electron from the metal).
Einstein, to whom this explanation of the laws of the photoelectric
effect belongs, went further and assumed that electromagnetic
radiation is not only absorbed and emitted in the form of quanta,
but is also propagated in that form.
Since the energy of a quantum is equal to Aco, and its velocity
is c, it should possess a momentum (see Sec. 21). It follows that
a quantum is a particle of zero mass. It will be noted that the energy
and momentum of an electromagnetic wave are related in exactly
this way (see Sec. 17).
The second phenomenon which exhibited the quantum properties
of radiation provided for confirmation of Einstein's hypothesis
concerning the momentum of light quanta. In the scattering of
Xradiation by electrons, the latter may be regarded as free, since
the characteristic frequencies of their motion in a substance are very
small compared with the frequency of the incident radiation (we
have considered such scattering in exercise 4, Sec. 19). It is essential
that in accordance with classical theory the scattered radiation
must be of the same frequency as the incident radiation. But exper
iment showed that the frequency of the scattered radiation is less
than the frequency of the incident radiation, and depends upon
the angle at which the scattering is observed (Compton effect). The
displaced frequency can be calculated in relation to the scattering
angle, if it is assumed that the act of scattering occurs as the col
lison of two free particles a moving quantum and an electron at
rest. A collision of this sort was considered in exercise 5, Sec. 21,
especially for the case of an incident particle with zero mass. The
equation obtained there gives a perfectly correct description of the
frequency shift in the Compton effect, if we consider that the energy
232 QUANTUM MECHANICS [Part III
of the quantum is equal to & = Aco and its momentum p = h k
(or h co/c in absolute magnitude).
Quantum mechanics. Thus, in the theory of light and the theory
of the atom, a peculiar dualism arose: one and the same physical
reality (the electromagnetic field or the atom) was described by
two contradictory theories: classical and quantum. A way out of
this situation was found in consistent quantum theory, where all
motion possesses certain wave properties, which, however, cannot
be detected in the motion of macroscopic bodies, but are essential
in the description of the motion of such microscopic particles as
quanta and electrons. The criterion to be used in order to determine
whether it is necessary to take into account the wave properties
of a given motion will be given in the following section. The only
thing to note here is that it involves the constant h.
The basic principles of quantum mechanics received direct experi
mental verification after the discovery of electron diffraction, whose
laws are very similar to the diffraction laws of electromagnetic waves.
All atomic phenomena are qualitatively and quantitatively fully
accounted for by quantum mechanics.
The present state of nuclear theory. Nuclear phenomena are some
what more complex. At the present time, we do not know the laws
governing the interaction of nuclear particles. These laws are closely
related to the properties of special nuclear fields, which properties
differ in many respects from those of the electromagnetic field. At
the present time, we still do not know the theory of nuclear fields.
It may also be that there is still insufficient experimental data for
the development of such a theory. Therefore, nuclear theory is, to
date, considerably less developed than atomic theory, in which all
interactions are of an electromagnetic nature and well known. In
any case, the difficulties experienced by modern nuclear theory
lie outside the region in which nonrelativistic quantum mechanics
can be applied, and in no way affect its basis.
The correspondence between geometrical optics and classical
mechanics. An essential role has been played in the formation of
quantum theory by analogy between classical mechanics and wave
optics. A correspondence will be established between them in the
present section. Since geometrical optics is a limiting case of wave
optics, an analogy between geometrical optics and classical mechanics
permits of transition to the wave equations of quantum mechanics
by means of a generalization. Let us establish an analogy between
the equations of mechanics and geometrical optics, quite formally
for the time being. Its meaning will be given later.
Surfaces of constant phase. Let us explain the significance of the
wave phase in geometrical optics. To do this, we perform the limiting
transition from wave optics to geometrical optics. We write the
expression for the field in the form
Sec. 22] THE ANALOGY BETWEEN MECHANICS AND GEOMETRICAL OPTICS 233
E = E (M)cos^^. (22.1)
A
Here, X is the wavelength, which is regarded as small compared with
the linear dimensions of the region occupied by the field. In the limiting
case of a plane wave the phase is
(22.2)
[Of. (17.21), (17.22)]. Since co = ^, k = ^> where u is the
phase velocity, it is convenient to obtain a relationship containing X
explicitly, describing the phase as 9 = y
The expression for the field must be substituted into the wave
equation
< 22  3 >
In differentiating with respect to i and r, we retain only those terms
containing the highest degree of X in the denominator because X is
a small quantity. Hence,
X
cos 
0** = ~ x dt* 0iiA x ~ x 2 \ dt I wo x
As stated, the first term in the last expression must be discarded
when X>0. Whence,
and similarly
_ x
=~~ COS T
obtain a firstorder differential equation for the phase 9 =  :
Substituting these expressions into the wave equation (22.3), we
2L
x
(22.4)
In the limiting case of a plane wave, it follows from (22.2) that
k = ^~V9 (22.5)
and
o>=~4f> < 22 ' 6 )
where
234
QUANTUM MECHANICS
[Part III
But, according to (22.4), this same equation is satisfied also by the
quantities ~ t jr in the expression of the almost plane wave (22.1).
It follows that we can take equations (22.5) and (22.6) as definitions
for the wave vector and frequency of an almost plane wave.
The wave vector is directed along the normal to a
surface of constant phase 9=^ const, i.e., along a
light ray at a given point of space. The propa
gation of an almost plane wave may be represented
as the displacement in space of a family of surfaces
of constant phase. In Fig. 35, these surfaces are
shown in cross section by solid lines, and the
light rays are dashed lines.
At various instants of time t, a surface of
definite value 9 = 90 occupies various positions in
space corresponding to the equation 9 (r, t) = 9 . We
determine the propagation velocity of this surface.
To do this, we proceed from the condition obtained by differentiating
9 = const:
Let dr be a vector in a direction normal to the surface. Then
is the absolute value of k. From (22.5) and (22.6), we obtain'
"5F
dt
"07
which coincides with the definition of phase velocity (18.7).
Phase and group velocities are, in general, different, since group
velocity is equal to
V = j. (22.7)
It is essential that for an almost plane wave, co may be expressed
as a function of k, just as is done in the case of a plane wave.
Surfaces of constant action. We shall now consider a family of tra
jectories of identical particles moving in some force field. For example,
these may be shrapnel particles formed when a shell is exploded
(though not pieces of the shell itself having different masses!); it
must be considered here that the shrapnel explosions occur at the
same place continuously, so that the particles fly one after the other
along each trajectory lagging only in time. It is not absolutely neces
sary to consider particles emerging from one point. Trajectories may
be taken which are normal to some initial surface. Each particle emerg
ing at t 1 has a definite trajectory depending on its initial coordinates
Sec. 22] THE ANALOGY BETWEEN MECHANICS AND GEOMETRICAL OPTICS 235
and initial velocity. The value of the action S for each particle may be
calculated along these trajectories from the equation [see (10.2)]
S^^Ldt. (22.8)
<0
Since at the instant t t the position of a particle is determined by
its trajectory, and L is a known function of coordinates, velocities,
and time, L{q(t), q(t), t}, the action along each trajectory is also
known as a function of time.
Let us join by surfaces the points of all the trajectories for which
the value of S is constant. The equation of this surface is S (r, t) = $ .
In accordance with (22.8), this surface coincides at the initial instant
of time with the surface from which the particles emerged.
The relationship of momentum and energy with action. The surfaces
of constant action are displaced in space and are orthogonal to the
particle trajectories, because, from (10.23),
__ dS _^_ _, ds
and, in general,
p = ==V$. (22.9)
r dT v '
The partial derivatives are proportional to the direction cosines of
the normal to the surface S=S Q .
Let us calculate, in addition, the partial derivative of S with respect
to time. Since the action depends upon the coordinates and the time,
its total derivative is equal to
BS = dS d^_ dr_ ,~2 IQ\
But from (22.8)
dS
Q Of
Substituting this into the lefthand side of (22.10), and p = = into
the righthand side of the same equation, we obtain
(22.11)
Similar quantities in an opticalmechanical analogy. Comparing
equations (22.9) and (22.11) with (22.5) and (22.6), we conclude
that a surface of constant action of a system of particles is propagated
similar to a surface of constant phase : the momentum of particles is
similar to the wave vector, while the particle energy is similar to the
236 QUANTUM MECHANICS [Part III
frequency. In accordance with Hamilton's equation (10.18), the particle
velocity v analogous to the group velocity of a wave is
(22.12)
The velocity of a surface of constant action is
88
u =
dt ~. (22.13)
This value by no means coincides with the particle velocity. Thus, for
free particles
a me 2 mv
so that u = . The quantity u is analogous to the phase velocity of the
waves.
The expression for group velocity (22.7) corresponds to Hamilton's
equation (22.12). In Sec. 18 it was shown that the group velocity corre
sponds to the velocity of propagation for a wave packet, i.e., a disturb
ance concentrated within a certain region of space. Thus, the analogy
between mechanics and geometrical optics establishes a correspondence
between a particle and a wave packet.
The transition to geometrical optics provides for representation of
the solution of the wave equation, in a certain region of space, in the
form of a plane wave ; however, the quantities defining this wave, such
as the frequency and the wave vector, are themselves slowly varying
functions of coordinates and time. The relationship between frequency
and wave vector will be of the form co = co (k, r), where the quantities
k and r that describe the wave propagation satisfy the same Hamilton
ian equations as p and r of a particle moving along a trajectory. It
is essential here that the vector k should not change much in magni
tude and direction over a distance of one wavelength, and that the
frequency co should not change greatly in one oscillation period.
G> = C& for a plane wave in free space; this is completely analogous
to the relationship (21.16) between the energy and momentum of a
particle of zero mass.
If light is propagated in an inhomogeneous medium, the phase
velocity u, appearing in equation (22.4), is a variable quantity. For
example, when light is refracted at the boundary of two media, u
has different values on both sides of the boundary. The propagation
of light in an inhomogeneous medium is similar to the motion of a
particle in a medium of variable potential energy.
C. 22] THE ANALOGY BETWEEN MECHANICS AND GKOMETBIOAL OPTICS 237
The opticalmechanical analogy was established by Hamilton in
>25. However, up to the time of the formation of quantum mechanics
e., up to 1925), the physical significance of this analogy was not
iderstood.
The law of transformation and the dimensions for similar quantities.
ie analogy between optical and mechanical quantities is relativisti
lly invariant. Comparing formulae (20.33)(20.36), for the trans
rmation of wave vector and frequency, with (21.18)(21.21), for
e transformation of momentum and energy, we see that similar
lantities are transformed in a similar manner.
The optical and mechanical quantities differ only in dimensions.
lus, phase has zero dimensions while action has the dimensions of
L dt, i.e., gm. cm 2 /sec. Accordingly, the wave vector and momentum,
id the frequency and energy, likewise differ by the dimensions of
tion. As can be seen from a comparison of the Lorentz transfor
ations for these quantities, the proportionality factor is an invariant
lantity.
In the following section we shall show that the analogy between
schanics and geometrical optics emerges as a limiting relation from
e precise wave equation of quantum mechanics.
Exercises
1) Formulate the equation of surfaces of constant action for a system
particles emerging from a single point in space x = 0, z = 0, in the plane
= in a gravitational field. The absolute value of the particle velocities
VQ and their direction is arbitrary.
We have
v x ^v x , x = mv X , x = v x t,
qt 2
iminating the initial conditions for velocities, we obtain
x dti mz mgt dS cy
    
ieed, from the expression for S we obtain
8S mUx\* lz\* g*tn mv*
*  IT ~ FIAT) + IT) + gz+ ~r\  y
2) Proceeding from the fact that phase is analogous to action, show that
ht of given frequency is propagated along trajectories for which the prop
ition time of constant phase is least (Format principle).
At constant frequency
f. , f ndr
9= kar = <o  .
it the product n di is equal to the displacement of the surface in a direction
rmal to it. It follows that  is the propagation time dt. In accordance
fch the variational principle, which governs phase as well as action, the time
ist be least.
238 QUANTUM MECHANICS [Part III
Sec. 23. Electron Diffraction
The essence of diffraction phenomena. Classical mechanics is anal
ogous only to geometrical optics and by no means to wave optics.
The difference between mechanics and wave optics is best of all illus
trated by the example of diffraction phenomena.
Let us consider the following experiment. Let there be a screen with
two small apertures. Let us assume that the distance between the
apertures is of the same order of magnitude as the apertures themselves.
Temporarily, we cover one of the apertures and direct a light wave
on the screen. We shall observe the wave passing through the aper
ture by the intensity distribution on a second screen situated behind
the first. Let us now cover the second aperture. The intensity distri
bution will be changed. Now let us open both apertures at once. An
intensity distribution will be obtained which will not in any way
represent the sum of the intensities due to each aperture separately.
At the points of the screen, at which the waves from both apertures
arrive in opposite phase, they will mutually cancel, while at those
points at which the phase for both apertures is the same, they will
reinforce each other. In other words, it is not the intensities of light,
i.e., the quadratic values, that are added, but the values of the fields
themselves.
This type of diffraction can occur only because the wave passes
through both apertures. Only then are definite phase differences ob
tained at points of the second screen for rays passing through each
aperture.
We disregard here the diffraction effects associated with the passage
through one aperture. These phenomena are due to the phase differ
ences of the rays passing through various points of the aperture.
Instead of examining such phase differences we consider that the
phase of a wave passing through each aperture is constant, but we
take into account the phase differences between waves passing through
different apertures. Nothing is essentially changed by this simplification.
Xray diffraction. In order to observe the diffraction of Xrays which
are of considerably shorter wavelength than visible light, they can be
made to scatter by correctly arranged atoms in a crystal lattice. Waves,
scattered by different planes of the lattice, have constant phase differ
ences 2 TU, 4 TC, . . . , etc., for definite scattering directions. The distance
between planes of the lattice, the scattering angles at which maxima
are observed, and the wavelength are related by a simple formula
(the WolfBragg condition). From this equation we can determine
the wavelength of Xradiation.
Diffraction by a crystal lattice is somewhat more complicated than
in the experiment with two apertures, though, fundamentally, it
occurs for the same reasons ; the wave is scattered by all the atoms of
the lattice, and the total amplitude of the scattered wave is the result
Sec. 23] ELECTRON DIFFRACTION 239
of adding all the amplitudes of the waves scattered by all the atoms,
with the path differences of the rays, i.e., the phase differences, taken
into account.
Electron diffraction. The very same phenomena is observed when
electrons (and also neutrons and other microparticles) are scattered
by crystals. As we know, electrons act on a photographic plate or a
luminescent screen in a way similar to Xrays ; as a result, direct exper
iment shows that microparticles undergo diffraction that is governed
by the same laws as the diffraction of electromagnetic waves.
However, for this, each electron must be scattered by all the atoms
of the lattice, because the electrons travel entirely independently of
each other; there can be no coherence (i.e., a constant phase difference)
between them, nor can any arise. They may even pass through the
crystal singly (see below). Only light waves which are originated from
the same light source exhibit diffraction in such a manner: a stable
diffraction pattern is obtained because the same wave passes through
both apertures. If the waves passing through the apertures originated
at different sources they could not cancel or reinforce each other at
fixed points of the screen. The alternation of light and dark regions
would depend on the relative phases with which the waves passed
through the apertures; a constant phase difference cannot be main
tained for light from different sources.
Electron diffraction demonstrates that the laws of motion in the
microworld are wavelike in character; to obtain the same diffraction
pattern for Xrays, each electron must be scattered by all the atoms of
the lattice. This is clearly incompatible with the concept of a definite
electron trajectory.
Diffraction phenomena prove that electron motion is associated with
a phase of a certain magnitude.
The de Broglie wavelength. From diffraction experiments we can,
without difficulty, determine the wavelength both for electrons and
for Xrays. For electrons, it turns out to be very simply related to
their velocities. Let us write down the expression for the wave vector
obtained from experiment:
k = . (23.1)
Here, A is a universal constant having the dimensions of action spoken
of in the preceding section (p. 229). It is equal to 1.054 x 10~ 27 ergsec.
Earlier, it was much more common to use a constant 2 n times bigger.
The value for h used in this book is frequently denoted by A.
The relation (23.1) was suggested in 1923 by L. de Broglie, before
the first experiments on electron diffraction. The quantity
27T _
A ~F
is called the de Broglie wavelength.
240 QUANTUM MECHANICS [Part III
Equation (23.2) shows that a certain wavelength can be ascribed
to the motion of each body, but in the motion of macroscopic bodies
it is extremely small as a result of the smallness of h compared with
the quantities which characterize the motion of macroscopic bodies,
where the orders of magnitude correspond to the cgs system. There
fore, diffraction phenomena do not actually restrict the applicability
of classical mechanics to macroscopic bodies.
The limits of applicability of classical concepts. The relationship
between the quantities is entirely different when equation (23.2)
is applied to the motion of an electron in an atom. The size of an atom
is determined, for example, from experiments on Xray diffraction or,
more simply, by dividing the volume of one gramatom of condensed
material (solid or liquid) by the Avogadro number N = 6.024 x 10 23 .
The atomic radius is of the order of 0.5 x 10~ 8 cm. From this it is
easy to evaluate the velocity of an electron by equating the "centri
fugal force" ^ to the force of attraction to the nucleus, equal to ^
in a hydrogen atom. For the velocity, the value obtained is
V f^f
whence the wavelength (23.2) is
27th 1/jrJ
A ~ e V m '
Substituting the numerical values m~10 27 gm, e~4.8 x io~ 10 CGSE,
we convince ourselves that the wavelength is approximately six times
larger than r. In other words, a distance of the order of an atomic
diameter can accommodate one third of a wave : this corresponds to
dimensions which are characteristic of diffraction phenomena and com
pletely "smears out" the trajectory over the atom.
Hence the motion of an electron in an atom is wave motion. Just
as the concept of a ray has no place in optics, if the light is propagated
in a region comparable with the wavelength, so the concept of an elec
tron trajectory becomes meaningless for the motion of an electron in
an atom.
The electron is not a wave, but a particle! It is necessary to warn
the reader against certain common delusions. First of all, contrary to
what is often written in the popular literature, the electron is never
a wave even in quantum considerations. Without any doubt the elec
tron remains a particle; for example, it is never possible to observe
part of an electron. If a photographic plate replaces the second (rear)
screen in the diffraction experiment, then at the point of incidence of
each electron there will appear a single point of blackening. The point
distribution characteristic of a diffraction pattern will also result when
Sec. 23] ELECTRON DIFFRACTION 241
the electrons pass through a crystal singly. * Thus, it is not the electron
that becomes a wave, but the laws of motion in the microworld that
are wavelike in character.
It is clear that a diffraction pattern can in no way be obtained from
a single electron. Since each electron gives a single point on the screen,
we must have very many separate points in order to obtain the correct
alternation of light and dark regions on the plate.
At the same time, diffraction would be utterly impossible if both
screen apertures or all the crystal atoms did not actually participate
in the passage of one and the same electron. In the diffraction experi
ment, electron trajectories simply do not exist. What is actually wave
like in the motion of a particle will be shown later.
In all its properties, the electron is a particle. Its mass and charge
always belong to it and are never divided in any diffraction experiment
or any other experiment that we know.
The incompatibility of trajectories and diffraction phenomena.
Another common delusion is that the electron supposedly does possess
a trajectory but that we are as yet unable to observe it due to imper
fections in technical facilities, or to the inadequacy of our physical
knowledge. In actual fact, the diffraction experiment shows that the
electron definitely does not have a trajectory, just as diffracted light
is not propagated in rays. To think that the development of physics
will in future show the existence of an electron trajectory in the atom
is just as unreasonable as to hope for the return of phlogiston in heat
theory, or of a geocentric world system in astronomy.
Statistical regularity and the individual experiment. The absence of
trajectories by no means signifies that we have lost all regularity. On
the contrary, an identical diffraction experiment, performed, of course,
with a large number of separate electrons of definite velocity, always
yields an identical diffraction pattern. Thus, causal regularity undoubt
edly exists. However, it is statistical in character appearing in a very
large number of separate experiments, because each electron passage
through a crystal may be regarded as a separate independent result.
Diffraction phenomena lead to a regular distribution of points on a
photographic plate in the same way that a large number of shots at
a target is subject to a law of dispersion. However, as opposed to bullets
which fly along trajectories and therefore give a smooth distribution
curve for the places where they strike the target, the blackened grains
on a photographic plate caused by electrons are produced in a more
intricate manner characteristic of wave motion. The distribution of bullet
fits is due to indeterminacy in the initial firing conditions and becomes
less in the case of better aiming, while the random character of electron
* In somewhat different form, this was shown by direct experiment by
V. A. Fabrikant, N. G. Sushkin and L. M. Biberman, using currents of very
low intensity.
16  0060
242 QUANTUM MECHANICS [Part III
behaviour presents a perfectly regular diffraction pattern and, for a
given electron velocity, can in no way be reduced.
In addition, it may be noted that the statistical regularity in a
diffraction experiment has nothing to do with the statistical regular
ities which govern the motion of a large ensemble of interacting par
ticles. As has already been repeated several times, the same pattern
is obtained completely independently of the way in which the elec
trons pass through the crystal all at once, or singly. A certain phase
governing the motion exists only because each electron interferes with
itself.
Electron trajectories in a Wilson cloud chamber. It still remains to
examine in more detail the question: In which cases do we, neverthe
less, deal with the concept of an electron trajectory ? In a cloud cham
ber, in a cathode ray oscilloscope, and in many other instruments,
electron trajectories can be precalculated very well from the laws of
classical mechanics. In a cloud chamber, there even remains a cloud
track along the line of motion of an electron.*
First, recall that under certain conditions light is also propagated
along definite trajectories (rays). Geometrical optics is applicable when
the inaccuracy in defining the wave vector Afc, subject to the inequal
ity
7r (23.3)
[see (18.9)], is small in comparison with k. Substituting Afc* for an
electron in equation (23.1), we obtain the analogous expression of
quantum theory:
(23.4)
This is the socalled uncertainty relation of quantum mechanics. The
concept of an electron trajectory has reasonable meaning if the un
certainties of all three momentum components kp x , A^y, A^ are
small compared with the momentum itself:
&p x <p x , &p y ^p y , &PZ<PZ. (23.5)
It may be pointed out that we have all along been saying c 'electron"
simply to be specific. The same applies to a proton, neutron, meson
and the like.
Let us suppose that the track of an electron in a cloud chamber is
0.01 cm wide and the electron energy is equal to 1.000 ev = 1.6 x 10~ 9
erg (1 electronvolt = 4 ' 8 = 1.6 X 10~ 12 erg). According to (23.4)
* An electron passing through a gas ionizes the atoms in its path. The
supersaturated vapour with which the chamber is filled condenses on the ions.
Upon illumination, the droplets appear in the form of a cloud track.
** Sometimes, Ap*, A# are meant to signify not the "spreadings" p x and x
in themselves, but their mean square values. Then, Ap x Aa? > h.
Sec. 23] ELECTRON DIFFRACTION 243
the component of momentum perpendicular to the trajectory has an
indeterminacy
6.6 x 10" ^ 66><10 _ 25>
and the momentum itself is
p = V2m $ = V2.9 x 10~ 36 = 1.7 x 10 18 .
It follows that the relationships (23.5) are satisfied to an accuracy
of up to four parts in ten million. The observation of a track in a
cloud chamber does not allow us to determine the trajectory with
accuracy sufficient to notice deviations (in electron motion) from the
laws of classical mechanics.
The limitations of the concepts of classical mechanics. Thus, quantum
mechanics does not abolish classical mechanics, but contains it as a
limiting case, much the same as wave optics includes the geometrical
optics of light rays as a limiting case. As we shall see later, quantum
mechanics is concerned with the same quantities as classical mechan
ics, i.e., energy, momentum, coordinates, moment. But the finiteness
of the quantum of action h imposes a limitation on the applicability
of any two classical concepts (for example, coordinates and momen
tum) for one and the same motion.
The coordinate and momentum of an electron cannot simultaneously
have precise values because the motion is wavelike. To attempt to
define these precise values is just as meaningless physically as it is
to seek precise trajectories for light rays in wave optics. In the same
way that it is impossible to obtain, as a result of improvements in
optical devices, a precise definition for light rays in wave optics, any
progress in measuring techniques as applied to the electron will not
allow us to determine its trajectory more precisely than indicated by
the relation (23.4), since strictly speaking the trajectory does not exist.
Attempts are sometimes made to interpret relation (23.4) errone
ously. It is taken that a trajectory cannot be determined because
the precision of the initial conditions does not exceed Ap* and A#
connected by relation (23.4). This would mean that some actual tra
jectory does exist but that it lies within a more or less narrow region
of space and within a certain range of momenta. The "real" trajectory
is likened to the imaginary trajectory from a gun to a target before
firing. The path of the bullet is not precisely known beforehand, if
only because strictly identical powder charges cannot be obtained.
But this inaccuracy in the initial conditions for the bullet only leads
to a smooth dispersion curve for the hits on the target, while the distri
bution of electrons indicates diffraction effects. The presence of dif
fraction shows that no * 'realthoughunknown tous" trajectory exists.
As a matter of fact, relation (23.4) by no means indicates with what
error certain quantities may be measured simultaneously, but to what
16*
244 QUANTUM MECHANICS [Part III
extent these quantities have precise meaning in the given motion.
It is this that the uncertainty principle of quantum mechanics expresses.
The term "uncertainty" emphasizes the fact that what we are con
cerned with is not accidental errors of measurement or the imperfection
of physical apparatus, but the fact of momentum and coordinate of
a particle being actually nonexistent in the same state.
Sec. 24. The Wave Equation
The wave function. Diffraction of light occurs because the wave
amplitudes are added. When the wave phases coincide the intensity
(which is proportional to the square of the resultant amplitude) is
maximum; when the phases are opposite the intensity is minimum.
In the diffraction of electrons, a quantity similar to intensity is meas
ured by the blackening of a plate, which blackening is proportional
to the number of electrons incident on unit area. The distribution of
the blackened grains on the plate obeys the same law as in the case
of the diffraction of Xrays (in the sense of alternation of maxima and
their relative positions). Thus, in order to explain the diffraction of
electrons we must assume that with their motion there can be associat
ed some wave function whose phase determines the diffraction pattern.
At the end of this section, we will show in the general case that such
a wave function must be complex, since a real wave function cannot
correspond to just any type of motion.
Probability density. Electrons move independently of one another
and pass through a crystal singly, as it were. Therefore, the number
of electrons in an element of volume dV is proportional to the proba
bility of the appearance of one electron. Probability (a quantity simi
lar to light intensity) must be quadratic with respect to the wave
function, in the same way that light intensity is quadratic in wave
amplitude. But since probability is a real quantity, it can only depend
on the square of the modulus of the wave function. Let us put
dw =  <p (x 9 y, z, t) \*dV. (24.1)
Here Aw is the probability of finding an electron in the volume dV
at the instant t ; then  <J>  2 is the probability referred to unit volume
or, otherwise, the probability density.
The linearity of the wave equation. Like in optics, where the laws
of propagation of the wave itself are studied on the basis of Maxwell's
equations, the intensity being found by squaring the wave amplitude,
it is necessary in quantum mechanics to find the equation governing
the quantity fy and not the probability density. This equation must
be linear. Indeed, two interfering waves, when combined, give a result
ant wave. In order to obtain the same interference pattern as in
optics, it is necessary to perform a simple algebraic addition of the
Sec. 24] THE WAVE EQUATION 245
functions ; both the summands and the sum must satisfy the same wave
equation. But only the solution of linear equations satisfy this require
ment. The phases of the waves are very essential here since in order to
formulate the laws of diffraction it is necessary to know not only the
behaviour of the squares of the amplitude but also that of their phases.
In other words, we must have an equation for the wave function
itself of the particle <Jj (x, y, z, t).
The wave function of a free particle. Proceeding from the analogy
between geometrical optics and mechanics, it is easy to construct a
wave corresponding to a free particle not subject to the action of
external forces. We know that the state of a free particle is character
ized by its momentum p. But, in accordance with relationship (23.1),
a wave with wave vector k = j corresponds to a particle with momen
tum p. It follows that the wave function of a free particle depends on
coordinates in the following way (we write it in complex form):
i P 
eiJ = e l ~h .
The time dependence of a wave function is also determined very
simply, if we recall that the frequency of a wave corresponds to the
particle energy (Sec. 22). The proportionality factor between them
has the dimensions of action. As was shown at the end of Sec. 22, the
factor of proportionality must be the same as between the wave vector
and momentum ; this follows from the condition of relativistic invar
iance of the correspondence between mechanical and optical quanti
ties.* Hence
<*) = f" < 24  2 )
Whence we obtain the wave function of a free particle
(24.3)
The group velocity of the waves is
v = f = ?f . (24.4)
dk dp v '
Hence, it coincides with the particle velocity as it should in accordance
with (22.12), thereby confirming (24.2).
The relation between wave function and action. We note that the
wave function (24.3) may be written in the form
ty = e IT , (24.5)
* Or at least from in variance to Galilean transformations.
246 QUANTUM MECHANICS [Part III
where 8 is the action of the particle. Indeed, the action of a free par
ticle is
&=**+pr, (24.6)
because from this we obtain
F dT v w, ^ a< ,
as should be the case according to equations (22.9) and (22.11).
Equation (24.5) confirms the relationship established also in Sec. 22
between the wave phase and the action of a particle.
The wave equation for a free particle. The analogy between mechanics
and optics by no means presupposes that the equations of mechanics
are written in a relativistically invariant form. It is sufficient to recall
that the analogy had already been established by Hamilton in 1825.
The significance of the analogy consists in the fact that a correspond
ence is established between quantities : momentum and wave vector,
energy and frequency, action and phase. In future, except in those
cases where it is specifically stated otherwise, we shall proceed from
the nonrelativistic form of the equations of mechanics.
Let us now find the differential equation satisfied by the wave func
tion (24.3). We have
**.= ",[,, (24.7)
_dj<___3_ ~
h
8x 2 '" (24.8)
d z fy p x
~dx* ^ W
Erom (24.7) and (24.8) we obtain
h * + &'
(here we are already using nonrelativistic expressions!), or in abbre
viated form
where A s the Laplacian operator. Equation (24.10) holds because
g =  , as can be seen from (24.9).
The Schrodinger equation. Let us now generalize equation (24.10)
to the case of a particle moving in an external potential field U. In
order to have a relation analogous to ^ = j h U since S = ~
for a free particle in (24.9), we must put
See. 24] THE WAVE EQUATION 24J
E. Schrodinger formulated this equation in 1925, generalizing the
de Broglie relations for free electrons to the case of bound electrons
(this was also before the discovery of electron diffraction).
Equation (24.11) directly follows from (24.10) for the simplest case
[7 = const, because then it is satisfied by the same substitution of
(24.3), though with the momentum value p = V~2m($ U) . From
here it is only one step to generalization to the case of variable poten
tial energy.
But this generalization must in no way be regarded as the deriva
tion of the Schrodinger equation from the equations of prequantum
physics, for it expresses a new physical law.
Its relationship to classical physics can be seen in the limiting tran
sition, which is fully analogous to the transition from wave optics
to geometrical optics.
The limiting transition to classical mechanics. Let us substitute the
expression for the wave function (24.5) in (24.11). This expression must
hold in the above limiting transition because then the wave phase
surfaces 9= const correspond to surfaces of constant action S = const
for particles.
9=X' (24.12)
Instead of the formal relationship considered in Sec. 22, we now
have an equality, since we have introduced a new, universal physical
constant h. Thus, we put
._$_
Whence
d<\i __ i dS ,
~df ~~ ~h~dT^ ''
^L * d# ,
i 8 Z S , L/ a>Sf
~h"dx*V ~~ fcMlte"
Let us substitute these derivatives in (24.11). Then, after eliminating
p, we obtain the equality
The limiting transition from quantum to classical mechanics is
attained by considering the de Broglie wavelength very small com
pared with the region in which motion occurs. Since the wavelength
is proportional to the quantum of action A, the same limiting transition
may be performed formally by considering that h tends to zero.
This means that all the quantities having the dimensions of action
248 QUANTUM MECHANICS [Part III
are so large compared with the quantum h, that the latter can be
neglected. In (24.13), passing to the limit & = 0, we have
a ' s ' _
~~~0T~ 2m l ~" *"'"'
or g = ^ + U from (22.9) and (22.11).
The limiting transition that we have performed here almost com
pletely repeats the transition from wave optics to geometrical optics
carried out in Sec. 22.
The correspondence between classical and quantum theory. We
have seen that the Schrodinger wave equation does in fact give a
correct limiting transition. This equation is, as it were, a fourth
member in the following correspondence:
geometrical optics > classical mechanics
I 1
wave optics > quantum mechanics I
The vertical arrows denote a transition from rays or trajectories
to wave patterns, while the horizontal arrows denote a transition
from waves to particles. The latter relates only to nonquantum
electrodynamics because in the transition to quantum field equations
the need arises for a corpuscular representation (see Sec. 27). Here,
we consider only the analogy between quantum mechanics and
classical wave optics.
The range of application of various theories. The regions in which
quantum mechanics and wave optics can be applied do not overlap
anywhere ; in wave optics or, what is just the same, in electrodynamics,
the velocity of light c is regarded as finite but the quantum of action h
is considered arbitrarily small. In nonrelativistic quantum mechanics,
c is considered arbitrarily large while h has a finite value. A quantum
theory of the electromagnetic field, in which both h and c have finite
values (i.e., the velocity ranges are comparable with c, and quantities
with the dimensions of action are comparable with h), has, in essentials,
also been completed at the present time. At any rate, any concrete
problem requiring the application of quantum electrodynamics, may
be uniquely solved to any required degree of precision, and the results
agree with experiment. The existence of a light quantum as an in
dependent particle is not a supplementary hypothesis which must
be made in order to formulate quantum electrodynamical equations.
The consistent quantization of electromagnetic field equations
necessarily leads to the corpuscular aspect of the theory (for more
detail see Sec. 27).
Sec. 24] THE WAVE EQUATION 249
The nonrelativistic particle quantum mechanics (i.e., constructed
on the relation $ = J~ + U) is, in the region for which it is applicable,
a theory which is just as consummate as Newtonian mechanics.
Like the equations of Newtonian mechanics, the wave equation
(24.11) is valid only for particle velocities small compared with the
velocity of light. But still, in the region for which it is applicable,
it is just as firmly established (in the same sense) as are the Newtonian
laws for the motion of macroscopic bodies.
The grounds for this are absolutely the same both nonrelativistic
quantum theory and Newtonian mechanics agree with the widest
range of experimental data, never contradicting them and providing
for correct and unique predictions. In addition, they nowhere contain
contradictory statements. The latter condition is, of course, not
sufficient for a physical theory to be correct but it is at least necessary.
The Bohr theory, or the old quantum mechanics (as it is otherwise
known) did not satisfy this requirement; in addition to the classical
concept of trajectory, it involved the quantum concept of discreteness
of states. For this reason, it had always been clear that the Bohr
formulation of quantum theory was not final and should be revised,
no matter how wide the range of experimental facts that it explained.
Quantum mechanics permitted the construction of a consistent
theory of atomic structure. The actual calculation of wave functions
for electrons in complex atoms is a problem of enormous mathematical
complexity.* However, it is, of course, by no means the purpose
of quantum mechanics to calculate the spectra of complex atoms:
the essential point is that quantum mechanics allows us to systematize
atomic and molecular states in such a way that the very nature
of the spectra is understood, whereas classical mechanics could not
explain even the stability of the atom. Thanks to quantum mechanics
such fundamentally important facts as the chemical affinity of atoms
or Mendeleyev's periodic law are now understood.
In its domain, quantum mechanics will, of course, perfect methods
of approach to various concrete problems. The correctness of its
general principles will serve as a basis for such refinement.
The normalization condition for a wave function. Let us return
to the wave equation (24.11). We shall write it for a wave function <p
and a conjugate function fy* (in the second equation we have to
replace i by i):
* It is considerably simplified thanks to approximation methods suggested
by V. A. Fok.
250 QUANTUM MECHANICS [Part III
Let us multiply the first equation by <**, and the second by fy,
and subtract the second from the first. The term ip* Ufy is eliminated
and the remaining terms give
 (24  15)
The lefthand side of the last equality is transformed to the form
_ fc_JLj;j, AJLm2
We can write the righthand side more fully thus:
h z h z
_ /^* ih d>Adf*) = (d>* divgrad A & divgrad d**) =
2971 2tTf\t
= ^ d* v (4 1 * grftd ^ ^ grad ^*)
[see (11.27)]. Finally, we represent the equality obtained in the
following form:
The lefthand side of this equality is the time derivative of the prob
ability density of finding a particle close to some point of space.
Let us integrate (24.16) over the whole volume in which the particle
might be situated. If this volume is finite then beyond its boundaries fy
and ij>* must be equal to zero. But then, from the GaussOstrogradsky
theorem
It follows that the integral itself does not depend upon time. It is
easy to see that it must be equal to unity because this is the probability
of an electron being somewhere, i.e., the probability of a trustworthy
event. The condition
J  ^2^7^! (24.18)
is called the normalization condition of a wave function.
If the electron motion is infinite, i.e., ip nowhere becomes zero,
the normalization condition appears more complicated. However,
in practice, it is always possible to consider that the volume in which
the electron is situated is very large and finite, so that the condition
(24.18) can be used. The physical results, naturally, do not depend
upon the arbitrary choice of volume.
Probabilityflux density. If we integrate (24.16) over any arbitrary
volume, we obtain
Sec. 24] THE WAVE EQUATION 251
(24 ' 19)
If on the left we have the change in the probability of finding
an electron inside the given volume, then on the righthand side
we must have the flux of the probability of it passing through the
boundary surface of the volume. According to (24.19) the density
rf the probability flux is equal to
It follows that a real wave function gives j = 0, i.e., it cannot
be used for describing the current of an electron. Therefore, in a general
lefinition, the wave function ^ must be a complex quantity.
The equation for stationary states. Let us assume that potential
energy does not depend explicitly upon time. Then classical mechanics
leads to conservation of energy of the system. The action of such
. o
a system involves a term St. But since fy =e h in quantum
mechanics, too, we must seek a wave function in the form
. fft
ty = e ' * <M*,y,z). (24.21)
Substituting this in (24.11) and omitting the zero subscript, we obtain
}he equation
#<1>. (24.22)
As we shall see in the next section, this equation has a solution
which does not satisfy definite necessary conditions for all values
of $. Thus, it turns out that, in contrast to the energy in classical
mechanics, the energy of a quantum system cannot always be
arbitrarily given.
Exercise
Prove that if there are two solutions of (24.22) for different values of energy
ff and <$", then
J
The functions ^* (r % &) and ^ (r, #') satisfy the equations

252
QUANTUM MECHANICS
[Part III
Let us multiply the first equation by ^> the second by t[*, and subtract the
second from the first. Integrating over the whole volume, similar to (24.19),
and then transforming the volume integral on the left to a surface integral,
we obtain the equation
_") f tt*(r,g)ty(r,&'
It follows that if S ^', then the second factor is equal to zero as required.
This is the socalled orthogonality property of wave functions. It will be shown
in more general form in Sec. 30 because it forms one of the most important
principles of quantum mechanics.
Sec. 25. Certain Problems of Quantum Mechanics
In this section we shall obtain solutions to the wave equation
for certain cases which are partly illustrative and partly auxiliary.
Nevertheless, many important laws are explained from these examples.
We have already obtained the solution of the wave equation for
a free particle (24.3). We shall now examine the solutions for bound
particles.
A particle in a onedimensional, infinitely deep potential well. Let
us suppose that a particle is constrained to move in one dimension
U remaining in an interval of length a, so
that Q^x^a. We can imagine that at
the points x = and x == a, there are ab
solutely impenetrable walls which reflect
the particle. A limitation of this type is
represented with the aid of the potential
energy curve shown in Fig. 36. U oo
at x<0 and x>a. We put U = Q at
0<#<a; this is the potential energy
gauge. To leave the region 0^x<a, a
particle would have to perform an in
finitely large quantity of work. Thus,
x
Fig. 36
the probability for the particle to be at x or x a is equal to zero.
With the aid of (24.1), we obtain
^(0) = &(a) = . (25.1)
These boundary conditions may also be justified by means of a limiting
transition from a well of finite depth. This will be done later.
Insofar as the potential energy is time independent, the wave
equation must be written in the form of (24.22). The motion is one
2
dimensional and, therefore, we must take the total derivative ^ T2
in place of A. From this we have
(26.2)
Sec. 25] CERTAIN PROBLEMS OF QUANTUM MECHANICS 253
We introduce the shortened notation
^^x, (25.3)
so that the wave equation will be of the form
*~ = x*ty. (25.4)
The solution to (25.4) is well known:
<p = C 1 sin Y.X + C 2 cos Y.X . (25.5)
But from (25.1) ip (0) = so that the cosine term must be omitted
by putting (7 2 = 0. There remains
<p = C I sin x x . (25.6)
We now substitute the second boundary condition
<Jj (a) = C l sin x a = . (25.7)
This is an equation in ^. It has an infinite number of solutions:
xa = (n+l)7r, (25.8)
where n is any integer equal to or greater than zero:
< n < oo . (25.9)
We discard the value n== 1 because, for n 1, the wave function
becomes zero everywhere, ^ = 8^0 = 0. Hence, <p 2 = 0, so that the
particle simply does not exist anywhere (a "trivial solution"). Now
substituting / from the definition (25.3) and solving (25.8) with respect
bo energy, we find an expression for the energy
^ h * >+l)2. (25.10)
Eigenvalues of energy and eigenlunctions. The boundary condition
imposed on a wave function is, for a given problem, just as necessary
as the wave equation itself. However, as can be seen from (25.10),
the boundary condition is not satisfied for all values of the energy,
but only for values belonging to a definite series of numbers by which
the problem under consideration is given. It will be seen later on
that, depending upon the conditions, these numbers may form a
discrete series or a continuous sequence. They are called eigenvalues
of the energy of a quantum mechanical system. The wave functions
belonging to the energy eigenvalues are called eigenf unctions.
The foregoing example is the simplest, one in which the energy
eigenvalues form a discrete series. The energy of a free particle forms
a continuous series of eigenvalues. Indeed, the only condition which
254 QUANTUM MECHANICS [Part III
can be imposed on the wave function of a free particle consists in
the fact that it must be finite everywhere, because the square of
its modulus is the probability density of finding the particle at a
given point in space. But the function (24.3) remains finite for all
real values of p and $.
The assembly of energy eigenvalues of a particle is termed its
energy spectrum. The energy spectrum in an infinitely deep potential
well is discrete, while the energy spectrum of a free particle is
continuous.
The solution of Schrodinger's equation (24.22) for stationary states
is always associated with finding the energy spectrum. In contrast
to the Bohr theory, where the discreteness of the states appeared
as a necessary but foreign appendage to classical motion, in quantum
mechanics the very nature of the motion determines the energy
spectrum. This will be seen especially clearly in the examples to follow.
The nodes or zeros of a wave function. The function ^ becomes zero n
times in the interval a (except at its ends). The quantity of
zeros ("nodes") of a wave function is equal to the number of the
subscript of the energy eigenvalue.
This result is easily understood from the following considerations.
In the interval a (for n = 0) there is one sinusoidal half wave ;
for n=l 9 there is one complete wave; for n = 2, there is one and a
half waves, etc. Thus, the greater n is, the less the de Broglie wave
length X. But energy is proportional to the square of the momentum,
i.e., it is inversely proportional to the square of X, in accordance
with (23.2). Therefore, the less X is, the greater the energy. This
conclusion holds, of course, for wave functions that are not of a purely
sinusoidal shape, though not as a general quantitative relationship
but, instead, qualitatively the more zeros or * 'nodes" that the wave
function has, the greater the energy. The least energy state has no
zeros anywhere except at the limits of the interval x Q and x~a.
It is called the ground state, all the other states being termed excited.
Normalization of a wave function. It remains to determine the co
efficient C l in order to define a wave function completely. We shall
find it from the normalization condition (24.18):
2
x sin2xa;\k__ C\a
)t
The second term of the integrated expression becomes zero at both
limits in accordance with (25.8). Thus,
SeC. 25] CERTAIN PROBLEMS OF QUANTUM MECHANICS 255
,yr
rin* (n+1) * . (25.12)
Real wave functions. The wave function (25.12) is real. Therefore,
from (24.20), the current in this state is equal to zero. This can also
be seen in the following way. The wave function (25.12) can be ex
panded into the sum of two exponentials. Each such exponential
represents, together with a time factor, the wave function of a free
particle (24.3), one of them corresponding to a momentum p = hx
and the other, to the same momentum but with opposite sign. Thus,
a state with wave function (25.12) is represented as the superposition
of two states with opposite momenta, these states having equal
amplitudes. The mean momentum for a particle moving in a potential
well according to classical mechanics is equal to zero; it changes
sign for every reflection from the walls of the well. In this sense,
we can say that the mean momentum is also equal to zero for quantum
motion. The difference is that at every given instant classical mo
mentum possesses a definite value, while the quantum momentum
of a particle in a well never has a definite value ; the wave function
involves states with momenta of both signs. This corresponds to
the uncertainty principle; since the particle coordinate is within
the limits 0<a;^a, the momentum cannot have an exact value.
In addition, we note that in this particular problem of a rectangular
well the square of the momentum has a definite value since the
uncertainty is extended only to the sign of the momentum. The
square of the momentum in this case is proportional to the energy.
The square of the momentum for a well of any arbitrary shape is
also not fixed.
A particle in a threedimensional infinitely deep potential well.
Let us now suppose that a particle is contained in a box whose edges
are a l9 a 2 , a 3 . Generalizing the boundary conditions (25.1), we conclude
that the wave function becomes zero on all the sides of the box:
= ty(x,a 2 ,z) = <l>(x,y,az) = Q. (25.13)
The wave equation must now be written in threedimensional form:
It is convenient to write the solution as follows:
4> = C f sinx 1 o; sinx 2 i/ sinx 3 z. (25.15)
It is written only in terms of sines and not cosines so as to satisfy
the first line of the boundary conditions (25.13). We substitute (25.15)
256 QUANTUM MECHANICS [Part III
in (15.14) and, utilizing the fact that for every factor of (25.15) there
exists an equality of the form,
~ sin Y.^X = x* sin K^X ; (25.16)
this gives
To satisfy equation (25.14) the energy must involve x x , x 2 and x 3
in the following way:
^ = JLL ( X j + X 2 + x j) . (25.17)
The quantities y. l9 x 2 , x 3 are determined from the second line of the
boundary conditions (25.13). The factors of (25.15) convert to zero
either at x = a l9 or y~a 2 or z a 3 . In other words,
sin x x i = , Xj a x
sinx 2 a 2 = 0, X 2 a 2 = 7i 2 7i: ; (25.18)
sin x 3 a 3 = , x 3 a 3 =
Here n l9 n 2 and n s are integers of which none are equal to zero (other
wise <p would be equal to zero over all the box).
Substituting x 1? x 2 , x 3 from (25.18) in (25.17), we have the energy
eigenvalues
7T 2 /? 2 ln\ nl n\\
=  ' 25 ' 19
The least possible energy is
'i=t(i + i + 5i) < 25  20 >
It follows that the value <^ = is impossible.
Calculating the number of possible states. To each value of the three
numbers n l9 n 2 , and n 3 , there corresponds a single particle state.
Let the numbers n l9 n 29 and n 3 be large in comparison with unity.
Such numbers may be differentiated: the differential dn denotes
a number interval which is small compared with n l9 but still including
many separate integral values of n . Then it stands to reason that
there are exactly dn : possible integers, l^dn^n^ included within
the interval dn^ (and similarly within the intervals dn 2 and dn 3 ).
Let us plot n l9 n 2 , and n 3 on coordinate axes. In this space we construct
an infinitely small parallelepiped of volume dn t dn 2 dn%. In accordance
with the foregoing, there are dn^ dn 2 dn 3 groups of three integers
%, n 2 , n 3 in this parallelepiped, each corresponding to one possible
state of the particle in the box. Altogether, the number of such
states in the examined interval of values n l9 n 2 and n 3 is
dN (n l9 n 29 n 3 ) = dn l dn z dn 3 . (25.21)
Sec. 25]
CERTAIN PROBLEMS OF QUANTUM MECHANICS
257
Substituting here x lf x 2 and x 3 from (25.18), we obtain another ex
pression for the number of states:
dN (x 1) x 2) x 3 ) =
where V a l a 2 a 3 is the volume of the box. The numbers x x , x 2 and
x 3 take only positive values.
It was pointed out above that to each value of x there correspond
two values of the momentum projection, which are equal in magnitude
and opposite in sign. Therefore, if we compare the number of states
included in the intervals dx ls and f == ~~ then there are half
ii n
as many states for the latter. Correspondingly, the number of states
in the interval of values of momentum dp x dp y dp z is
dN ( Px , ft, ft) =
(25.23)
where p X) p y and p x assume all real values from oo to oo.
Equation (25.23) agrees with the uncertainty relation (23.4). If
the motion is bounded along x by the interval a l9 then only those
states differ physically for which the momentum projections differ
by not less than TC . Hence, there are ^ rr~r = ^ ? states
J ! ' (2nkla^ 2nh
within the interval dp x . Multiplying r P r a ~T~ * 2 f* we arr * ve
at (25.23). In order to ensure coincidence of numerical coefficients
with the results of rigorous derivation from the wave equation when
evaluating the number of states from
the uncertainty relation the quantity
2nh was selected on the righthand side
of (23.4) or 2n from (18.10).
We shall now consider the number of
states after changing somewhat the inde
pendent variables. We plot the quantities
x 1? x 2 and x 3 on the coordinate axes
(Fig. 37). Let us construct in this "space"
a sphere whose equation looks like
The numbers x 5 x and
x 3 are posi
Fig. 37
tive so that we shall be interested only
in one eighth of the sphere; this octant is shown in Fig. 37. How
many states are included between the octants of two spheres with
radii K and K + dK? The number of states is equal to the integral
of (25.22) over the whole volume between the octants, or
17 0060
258 QUANTUM MECHANICS [Part III
f
S5 J
x .7t /or 0/n
f x 2 , x s ) =  3  =  2 ' ( 25  24 )
This is evident simply from the fact that the volume is equal to the
4 172
surface of the octant ^p multiplied by dK. But from (25.17) K
is very simply related to the energy of the particle:
Whence
' (2525)
Thus, the number of states included between and ${d$ in
creases in direct proportion to t^ 1 / 2 . In a onedimensional potential
well we would obtain dN = dn= * = %r? ! IPI T Equation (25.25)
a V '
has great significance in all that is to follow. We observe that this
formula involves only the volume of the vessel F, irrespective of
the ratio of the edges a l9 a 2 , and a 3 . In mathematical physics courses
it is shown that the result (25.25) holds for energy eigenvalues which
are sufficiently large compared with the energy of the ground state.
The number of states is proportional to the volume of the vessel
and is independent of its shape.
A onedimensional potential well of finite depth. We shall now con
sider a onedimensional potential well of finite depth. We specify
it in the following way : U = oo at oo <
<a<0, U = for O^x^a and U=U
for a<x^. oo. In other words, the potential
energy for x> is everywhere equal to J7 ,
except within a region of width a near the
coordinate origin, which region we called
the well. For x<0 the potential energy is
a J infinite (see Fig. 38).*
Fig. 38 Since the solution will be of different
analytical form inside and outside the
well, we must find the conjugation conditions for the wave func
tion at the boundary.
Let us take the wave equation.
79 JO I
(25.26)
* It was shown in Sec. 19 that the threedimensional wave equation can
be reduced to a singledimensional one, with the difference that the variable r
must be positive by its very meaning. This may be attained formally by situating
an infinitely high potential wall at r = 0. Fig. 38 actually refers to a spherical
potential well with an angularmomentum value equal to zero, when there
is no "centrifugal" term in the potential energy [see (5.8) and (31.5)].
Sec. 25] CERTAIN PROBLEMS OF QUANTUM MECHANICS 259
and integrate both sides over a narrow region a 8<#< a +8,
including the point of discontinuity of the potential energy x = a.
The integration gives
a+8
[(4.) (4*) lf(/Z7)+fa. (25.27)
2 m l\ dx I a+ 8 \dx I a _8 ] J v ' T v '
a 8
Even though 7 suffers a discontinuity at the boundary of the well,
on the right it remains everywhere finite by arrangement. Therefore,
when 8 approaches zero, the integral on the right also approaches
zero. It follows that the lefthand side of (25.27) is also zero. In other
words,
(*+.) (<+.) . (26.28)
\dxl a +o \dxlaQ' '
the limit of the derivative on the right is equal to its limit on the left.
This argument would not hold in the problem of an infinitely
deep well because then the integral in (25.27) would be indeterminate.
Besides we notice that the derivative ^ is finite at the points x =a 8
simply because the only solutions of equation (25.26) are those with
finite derivatives (exponential function, sine or cosine).
We shall now show, by means of a limiting transition, that even
the wave function itself does not suffer a discontinuity at the boundary.
Let U initially have a finite discontinuity region of width 8 and let
the discontinuity of the function fy be A. Before passing to the limit
the derivative in the region of discontinuity is of the order of ~ ~ ,
so that when 8>0, it diverges. Let us now multiply both sides of
(25.26) by fy and perform a transformation by parts:
Let us integrate the transformed expression between a 8 and
a + (5. We then obtain
a+8 a+8
a 8 a 8
(25.59)
We shall now perform the foregoing limiting process by indirect proof.
We may write the integrated terms thus:
a+8
because the derivative r^, as was shown, is not subject to a dis
continuity. Within the assumed discontinuity region of the ^function,
260 QUANTUM MECHANICS [Part III
r^ is of the order of f , but at the boundaries of the region it reverts
dx o
to values which are independent of 8 and are therefore finite in the
limit. Hence, the whole integrated part on the left in (25.29) is of the
order of A jrH The remaining integral is estimated as
Hence it tends to infinity as S tends to zero. The righthand side of
(25.29) is finite for >0. Thus, by assuming that fy has a finite dis
continuity A we have arrived at a contradiction. It follows that fy
is continuous at the point a together with its first derivative.
Solutions in two regions. The wave equation for the region
(inside the well) is of the form
We take its solution
(25.30)
where x is defined from (25.3). The solution involving the sine only
is taken because at the lefthand edge of the well, where the potential
energy suffers an infinite discontinuity, fy satisfies the boundary
condition (25.1): 9(0) = 0.
The wave equation outside the well, when x>a, is
IT^. (25.31)
0/ T v '
__ .
2m dx 2 v 0/ T
First of all let us take the case > >U . Then, introducing the ab
breviated notation
*?!(*_ ff o)53X , (25.32)
we obtain (25.31) in the standard form (25.4)
whence
fy = (7 2 sin x x x + (7 3 cos x x x . (25.33)
We must now satisfy the boundary conditions on the righthand
edge of the potential well where U suffers only a finite discontinuity.
According to these conditions the wave function is itself continuous, i.e. ,
C l sin xa = C 2 sin X A a + O 3 cos Xj a (25.34)
and its derivative
x G! cos xa = x x C z cos ^ a x x (7 3 sin K t a . (25.35)
Sec. 25] CERTAIN PROBLEMS OF QUANTUM MECHANICS 261
From these equations we can determine (7 2 and (7 3 in terms of C v
i. e., completely express the solution outside the well in terms of the
solution inside the well. The equations (25.34) and (25.35) are linear
with respect to (7 2 and G 3 and have solutions for all values of coefficient.
n x 1 sinxasinx 1 a+
== 
Therefore, the boundary conditions may be satisfied for any real values
of x and x r Thus, Schrodinger's equation is solvable for all &. There
is no discrete energy eigenvalues for $ > U .
We could adjust the potential energy in this problem to zero at
infinity, i.e., consider it equal to zero for x > a and equal to U
for a>#^0. Then the case which we have just considered would
correspond to positive eigenvalues of the total energy.
Now let $ < U Q . We introduce the quantity
(25.36)
The wave equation is now written differently from that for $ > C7 ,
namely
d*<b ~ 9 .
~ = X 2 U>.
dx* T
Its solution is expressed in terms of the exponential function
<J, = <7 4 e^ + 5 e Sr . (25.37)
But the exponential eZx tends to infinity as x increases. For x == oo it
would give an infinite probability for finding the particle, and no
00
finite value could be assigned to the integral f  fy  2 Ax . It follows that
o
a physically meaningful solution exists only for 64 = and must be
of the form
ty^Cse (25.38)
Let us again try to satisfy the boundary conditions at x = a. This
time they appear as follows:
C^sinxa = (7 5 e~ xa , (25.39)
x C^cos xa = x C 5 e*' . (25.40)
Let us divide equation (25.40) by (25.39) in order to eliminate C l
and (7 5 . We then obtain
x cot xa = x . (25.41)
262
QUANTUM MECHANICS
[Part III
From this equation we find the expression for sin xa :
1 . 1
sin xa ==
_i
cot a xa
.._ _ i
(25.42)
Let us reduce this equation to a more convenient form. From (26.3)
. xa ,
so that
smxa = 
. xa,
(25.43)
only those solutions should be chosen for which ctg xa is negative, in
accordance with (25.41), i.e., xa must lie in the second, fourth, sixth,
eighth, etc., quadrants.
We shall solve this equation graphically (Fig. 39). The lefthand side
of equation (25.43) is repre
sented by a sinusoid, while
the righthand side is repre
sented by two straight lines
of slopes 7=^== . If
the absolute value of the
slopes of the angle of incli
nation of these lines is less
than 2/Tu, they have one or
several common points with
Fig. 39 the sinusoid in the quadrants
corresponding to the roots
of (25.41). The trivial point of intersection xa = does not count
because, for x = 0, the wave function is zero everywhere. Thus, in a
well of finite depth of the form considered, there are only several
energy eigenvalues.
If
8 ma 2 '
there are in general no points of intersection of the straight lines with
the sinusoid corresponding to energy eigenvalues. In Fig. 39 the
points of intersection in the even quadrants are marked by small
circles.
SeC. 25] CERTAIN PROBLEMS OP QUANTUM MECHANICS 263
Finite and infinite motion. We shall now relate the shape of the
energy spectrum to the type of motion. For $ > U Q the solution
outside the well is of the form (25.33). It remains finite also for an
infinitely large x. Therefore, the integral J  fy  2 dx taken over the
region of the well is infinitesimal compared with the same integral
taken over all space. In other words, there is nothing to prevent the
particle going to infinity. Such motion was termed infinite in Sec. 5.
For <f < C7 , the solution (25.38), if it exists, is exponentially damped
at infinity. Hence, the probability of the particle receding an infinite
distance from the origin is equal to zero the particle remains at a
finite distance from the well all the time. This motion was termed
finite in Sec. 5.
Thus, infinite motion has a continuous energy spectrum while finite
motion has a discrete spectrum consisting of separate values. If the
depth of the well is very small, the finite motion may be absent. It has
no counterpart in classical mechanics. Finite motion is always possible
in a potential well if  $ \ < U.
The result that we have obtained does not only refer to a rectangular
potential well. Indeed, if the potential energy is taken to be zero at
infinity, then the solution with positive total energy is of the form
(25.33) for sufficiently large x, while the solution with negative total
energy is of the form (25.38). The latter contains only one arbitrary
constant while (25.33) contains two constants. Both solutions must be
extended to the coordinate origin in order that the condition fy (0) =
can be satisfied at the origin (we consider that x is always greater
than zero). Obviously, if we have two constants at our disposal we
can always choose them so that the condition fy (0) = is satisfied. *
Contrarily, a solution of the form (25.38) containing one constant
becomes zero at the origin only for certain special values of x.
A continuous spectrum corresponding to infinite motion may be
accounted for in the following way. A free particle moving in unbound
ed space has a continuous spectrum. The wave function of the particle
in infinite motion differs from the wave function of a free particle
only in the region of a potential well. But the probability of finding
the particle in this region is infinitesimal if the whole region of motion
is sufficiently large. Therefore, the wave function for infinite motion
coincides with the wave function of a free particle in * "almost" the
entire space, i.e., in that region of space for which the probability
of finding the particle is equal to unity, and the energy spectrum
turns out to be the same as for a free particle.
The wave function in a region where the potential energy is greater
than the total energy. If U Q tends to infinity the function outside the
* If *(0) = 0^(0) + C7 2 <MO) , then ^ =  J^.
264 QUANTUM MECHANICS [Part III
well very rapidly tends to zero. In the limit C7  oo, it tends to zero
however close to the boundary # a, thereby giving the boundary
condition (25.1).
In the case of a finite U the wave function outside the well does
not become zero at once. Therefore a finite probability exists that
the particle will be outside the well at a finite distance from it.
This would have been completely impossible in classical mechanics,
as is obtained from (25.38) in the limiting transition h . >0 for x = co
and tp however small outside the well. This, naturally, should be the
case : if the particle is situated outside the well its kinetic energy is
$ U Q <Q. But the velocity of such a particle is an imaginary
quantity. In classical mechanics it means that a given point of space
is absolutely unattainable for the particle at the given value of its
energy <^. In quantum mechanics, a coordinate and velocity never
exist in the same states as precise quantities. Earlier, we interpreted
this in terms of the uncertainty relation, i. e., we considered cases for
which precision in the concept of velocity for a certain state was
restricted by the limits  ^ . However, this is a lower limit and has
to do with particles which are almost unaffected by forces. The appear
ance of an imaginary velocity in the equation for a bound particle
shows that the very concept of velocity is not applicable to a region
of space, however large, for which U > $ . We can express this diifer
ently by saying that, for U >$, the uncertainty in the kinetic energy
is always greater than the difference U $.
To summarize, in classical mechanics there is no counterpart to
the motion of a bound particle outside a well.
Exercises
1) The potential energy is equal to zero for x < and equal to J7 for x>
(the potential threshold). Incident from the left are particles with energies
& > (70. Find the reflection coefficient.
The wave function on the left is
 _.
C\e h + C 2 e h .
On the right, above the threshold, the function is
C 3 e * , where p' \/2m (& U) .
Find the ratio IC 2 ) 2 /!^! 2 from the boundary conditions at a? = 0, i.e., the
ratio of the squares of the amplitudes of the reflected and incident waves.
The ratio is equal to unity for & < U .
2) The potential energy is equal to zero for x < and for x > a. U = U Q
for 0<a?<a (a potential barrier). Particles are incident from the left with
energies less than C7 . Find the coefficient of reflection.
Sec. 26] HARMONIC OSCILLATORY MOTION 265
The wave function to the left of the barrier is equal to e* kx f Ce~ ikx I k ~ 1 ;
under the barrier, i.e., for 0<#<a, the wave function is C&** C&**. We
look for a wave function of the form C^ kx beyond the barrier. This means
that beyond the barrier the only wave is that travelling to the right (i.e.,
only the transmitted wave), while in front of the barrier wo find both the incident
wave and the reflected wave. The constants C, C 19 G 2t C7 3 are determined from
the continuity condition of the wave function and its first derivative at the
boundaries of the barrier. The expressions for the constants C and (7 3 are as
follows :
2(x 2 + fc 2 )shxa
~~ (x f ik) 2 e *<* ~ (x
(x + ik) 2 e *<* (x i
The particle flux on the left and right of the barrier is, respectively,
Substituting C and C 3 it is easy to see that both the expressions for flux coincide
as expected.
If xa> 1, i.e., the barrier is transparent to a very small extent, we have
C  1, C 3 =
x
Thus the flux diminishes exponentially with the thickness of the barrier.
It will also be noted that the total particle flux through the barrier is pro
portional to the particle density in front of the barrier, because the boundary
conditions are linear and homogeneous with respect to the wave functions.
By specifying the amplitude of the wave function on the left we determined
the density and flux of the particles.
3) Verify the orthogonality property for wave functions (the exercise
in the preceding section) for a particle in a box of finite and infinite depth.
Sec. 26. Harmonic Oscillatory Motion in Quantum Mechanics
(Linear Harmonic Oscillator)
The wave equation lor an oscillator. In Sec. 7 we considered har
monic oscillations with one degree of freedom. The Hamiltonian func
tion of this system, called a linear harmonic oscillator, is of the form
Forming Hamilton's equations, we obtain
, ^
* dx ' dp
Eliminating p y we arrive at the usual equation of harmonic oscillations
(7.13):
266 QUANTUM MECHANICS [Part III
In quantum mechanics the wave equation corresponding to this
motion has the form [see (24.22)]
 + J5f!! *'+ < 26 ' 2 >
Indeed, since the motion has only one degree of freedom, instead of
the Laplacian A, we must simply write the second derivative. The
, . , . T . mco 2 cc a
potential energy is equal to  .
Let us now introduce other units of measurement, in particular,
we shall take the unit of length equal to / l , so that
^ v mco '
x = 1/1515. (26.3)
\ mco ^ v '
The quantity is dimensionless. The derivative ? is equal to
, = .. (26.4)
dx \ h d% v '
Further, we put
2<? = 7ico. (26.5)
In terms of these dimensionless variables, equation (26.2) assumes
the form
_.!*! + 52 <j, = e <,. (26.6)
Equation (26.6) does not contain any parameters of the problem,
i.e., co, m, and h. For this reason the eigenvalue s can only be an
abstract number. Comparing this with the expression for energy
(26.5), we see that the energy eigenvalue of an harmonic oscillator is
proportional to its frequency co.
The transition to another dependent variable. It appears convenient
to introduce a new dependent variable:
(  ) = e Ty (5) . (26.7)
Whence
(26.8)
~
We substitute (26.8) in (26.6) and perform the necessary rearrange
ment. The new dependent variable g () having been introduced, the
equation assumes the form
Sec. 26] HARMONIC OSCILLATORY MOTION 267
(26.9)
Integration in the form of a series. It is possible to integrate equation
(26.9) by expanding it in a power series of the form:
00
g (?) = So + 9! I + 9* ? + ft ? + =2>$" (2610)
n0
In order to determine the coefficients of the expansion g n , we must
substitute the series (26.10) into equation (26.9), differentiate it by
terms and compare the expressions for the same powers of 5. The
first derivative is
so that
The second derivative is
oo
*. (26.12)
In the last summation we changed the summation index, denoting
it by the letter k. We shall now revert to n, assuming k 2 = n,
. Then
+ 2) (n + 1)^,5". (26.13)
Now substituting (26.13) and (26.11) into equation (26.9) and collecting
coefficients of n , we obtain
00
[(n + 2) (n + l)7 M+2 + 2njr (e l)ffj  . (26.14)
We know that for a power series to be equal to zero, all its coefficients
must convert to zero. Thus,
(7i) (26 ' 15)
In this way the expansion proceeds in powers of (5 2 ) because the
coefficients g n go alternately.
Examining the series. Let us assume initially that g Q ^Q. Then,
from equation (26.13), we find in turn gr 2 , gr 4 , . . ., g z k. Not a single odd
268 QUANTUM MECHANICS [Part III
coefficient will appear in the series if g = 0. On the contrary, if g Q = 0,
(/^O, then no even coefficients will appear in the series; for this
reason it is sufficient to examine solutions containing only even or
only odd powers of . To be specific let us first take the series in even
powers.
Let us examine the behaviour of the series (26.10) for large values
of . Terms involving high powers of 5, i. e., large n are then predomi
nant. But if n is a large number then, in equation (26.15), we can neg
lect all constant numbers where they appear in the sum or, difference
with n. If n is large the equation will take on the form
g, + i = 2gn. (26.16)
IV
Let n = 2 n 1 so that ri now changes by unity only. Putting this in
(26.16), we obtain
gi> + i=rg'*; (26.17)
where we have introduced the notation g' n ' =s g 2n * ===g n \g' n > are coeffi
cients of the series in powers of ( 2 ).
If we now take a function containing only odd powers of , then
the terms involving g^ n f \i will n ^ differ, for large n', from the
terms of a series in even powers of because unity can be neglected
in comparison with 2 ri . Therefore, the form of the coefficients for
large ri is identical both for series in even and odd powers of .
From (26.17) we find an expression for g' n '\\\
Whence the expression for gfe) in the case of large is of the form:
 = &* < 2<U9 >
n'  n'
Thus, the asymptotic expression for g(%) is the exponential function
e* 2 . But then, in accordance with the definition of g(fy (26.7), the
asymptotic form of ^() for large 5 * s
= 2
However, this form of ^ is not acceptable : the wave function must
remain finite at infinity because its square is probability.
The condition for eigenvalues. There is only one possibility for
obtaining a finite value of fy at infinity. It is necessary that the series
(26.10) should terminate at a certain n and that all the subsequent
Sec. 26] HABMONIC OSCILLATORY MOTION 269
coefficients <7 + 2 <7f4> etc., be identically equal to zero. It can be
seen from equation (26.15) that g n + 2 becomes zero when
==271+1, (26.20)
where n is any integer or zero. Since g n + t is linearly expressed in
terms of g n + 2 > it is sufficient for g n + 2 to convert to zero to have the
series terminate at g n . It follows that when e satisfies (26.20) the
function g(%) becomes a polynomial. The product of the polynomial
a
0() with the exponential 2" always tends to zero as tends to
infinity. Hence ^ (oo) = 0. As was pointed out at the end of the last
section, such motion is finite in the same sense as in classical mechan
ics : the probability that the particle will recede an infinite distance
is equal to zero. To finite motion there corresponds a discrete energy
spectrum; from (26.5) and (26.20)
* B = ^e = *fi>(+y). (26.21)
The least possible value of energy 6\ = " * As we have already
said in Sec. 25 a state with energy <^ is called the ground state. For
this state the series gfe) is already terminated at the zero term, because
the number of the eigenvalue of energy determines the degree of the
polynomial gn(>). The wave function of the ground state is of the
especially simple form:
~
This function does not have any zeros at a finite distance from the
coordinate origin, which must be the case in the ground state. It may
be noted that the state with zero energy would correspond to a par
ticle at rest at the origin. However, such a state is not compatible with
the uncertainty principle, since it has, simultaneously, a coordinate
and velocity.
Oscillator wave functions. Let us also find the eigenfuiictions for the
first and second excited states. In the first state ff l Aco ( 1 f ^\ =
3
= ^ A co. The series will be terminated if we assume <7 = 0, ffi^*
A
Then e = 3 from (26.20) and gr 3 =0r 5 =0r 2w + 1 = 0, etc., from (26.15).
The even coefficients may be assumed at once equal to zero, for which
purpose it is sufficient to take gr = 0.* In general, all functions with
even n turn out even, while those with odd n are odd. In accordance
with what has just been said, the wave function with n = 1 is
* If gr ^ for e = 3, then the series in even powers of would have extended
to infinity, which, as was shown, is impossible.
270
QUANTUM MECHANICS
[Part III
2 . (26.23)
This function becomes zero for = 0, i.e., it has one node.
In the same way it is easy to find ip 2  Indeed, <^ 2 A co 12 + ^1 =
= ~&G), s==5. From (26.15), the coefficient </ 2 is
so that
(26.24)
(26.25)
1
The nodes of this function are situated at the points s db ^ 
In general, the function fy n has n nodes. The functions for several
small values of n are shown in Fig. 40.
We show the eigenvalue distribution and the potential energy
curve in Fig. 41. It is very curious that the
eigenvalues are separated by equal intervals.
The oscillator problem qualitatively resembles
Fig. 40
Fig. 41
that of an infinitely deep rectangular well (Sec. 25), but the energy
level in the well increases in proportion to the square of the number.
Exercises
1) Show that, neglecting a constant factor, the function g n ( ) may be written
in the form
ft,ra.*/ dVl
'/AY 1
U/
Verify this by substitution in equation (26.9), in which e = 2?Hl.
2) Normalize the functions % and ^, taking advantage of the fact that
 00
(see exercises of Sec. 39).
Sec. 27] QUANTIZATION OP THE ELECTROMAGNETIC FIELD 271
Sec. 27. Quantization of the Electromagnetic Field
The electromagnetic field as a mechanical system. An electromagnetic
field in a vacuum may be regarded as a mechanical system ; this was
shown in Sec. 13. It possesses a Lagrangian function, action, and so
on. We are, therefore, justified in posing the problem of quantization
of this system, i.e., applying quantum mechanics to it.
The basic difference between electrodynamics and the mechanics
of point masses is that the degrees of freedom of an electromagnetic
field are distributed continuously: in order to specify the field at a
given instant of time, we must define its value at every point of space.
In this sense electrodynamics resembles the mechanics of liquids or
elastic bodies, if one regards them as continuous media ignoring the
atomic structure of the substance. The degrees of freedom of a field
are labelled by the coordinates of points in space, while the amplitude
values of the potential are generalized coordinates [see (13.2)]. Poten
tials are usually chosen as generalized coordinates because they satisfy
secondorder equations in time, as do generalized coordinates in
mechanics.
The potentials satisfy the Lorentz condition, which reduces to
div A = 0, provided the gauge transformations are chosen so as to
eliminate the scalar potential.
The electromagnetic field coordinates defined in this way are not
independent of one another. Indeed, the equations of electrodynamics
involve coordinate derivatives, i.e., differences of field values at
infinitely close points. In this sense, field equations resemble the
equations for coupled oscillations : they are linear, but each one in
volves several generalized coordinates instead of one. The equations for
coupled oscillations can be reduced to normal coordinates which are
mutually independent. The same can be done with the wave equations
of electrodynamics, thus separating the dependent variables therein.
This considerably simplifies the application of quantum mechanics
to radiation.
Clearly shown here is the generality of the methods of analytical
mechanics: they permit determining generalized coordinates and
momenta in such manner that quantum laws can then be applied
uniquely.
The electromagnetic field in a closed volume. We must first of all
represent the electromagnetic field as some kind of closed system,
since quantum mechanics is most conveniently applied to such
systems. We can assume, for example, that the radiation is contained
in a box with mirrortype reflecting walls. At the walls of such an
imaginary box (# = or xa^ y = Q or y a 2 , z = or z=a 3 ) the
normal components of the Poynting vector TJ become zero. However,
it is simpler to suppose that the field is periodic in space, and the lengths
of the periods in three perpendicular directions are equal to the di
272 QUANTUM MECHANICS [Part III
mensions of the box; the period of the field along x is equal to a l9
that along y is equal to 2 > an ^ that along z is equal to 3 . In other
words,
A (x, y,z)=A(x + a lt y, z) = A (x> y + a 2 , z) ==
= A(*,y,3 + a,). (27.1)
We have, as it were, divided space into physically identical regions,
after which it is sufficient to consider a single region.
The solution of equations describing a harmonic field in free space
was found in Sec. 17 [see (17.21)]. Introducing a time dependence into
the amplitude factor, we represent the potential in the following form :
A (k, r, t) = A k (t)e** + A* (t) e~'* , (27.2)
where its reality is shown explicitly.
The potential satisfies the Lorentz condition which, for a plane
wave, can be reduced to the form divA = (since 9 = 0); whence,
by (11.27), we obtain
'* kr ) + div(A*e' kr ) ==
= (A k Ve {kr ) 4 (A* Ve' to )
In order that this equation be satisfied for all r, the coefficient of
each exponential term must convert to zero. In other words, the
vectors A k and At are perpendicular to the wave vector k:
(kA k )=0, (kAy = 0. (27.3)
For each k there exist two mutually perpendicular vectors A (or = 1,2 )
corresponding to two possible wave polarizations. Any vector in a
plane perpendicular to k can be resolved into A k 1} and A k 2) .
We shall now apply the periodicity condition (27.1) to each term of
(27.2) separately. For the first term we obtain
= A
whence it follows that
Therefore the components of the wave vector should be
L 2TCn 2 Tf * TC 3 /97 A\
A/y , tt/z , \i I .Tty
where n lf n 2 , n 3 are integers of any sign.
Consequently, each harmonic oscillation is given by three integers
^i> ^2 % an( i a polarization CT, which can take two values. As indicated
in Sec. 13, A lrtaWs is a generalized coordinate. The number of such
Sec. 27] QUANTIZATION OF THE ELECTBOMAGNETIC FIELD 273
coordinates is infinite, but at least forms a denumerable set, and not a
continuous set equivalent to the set of all points in space.
This, then, is the basic simplification introduced by the periodicity
condition. This condition is, of course, only a mathematical con
venience, there being no basic periods a x , a 2 , and 3 in any final
physical result.
An electromagnetic field is specified if its oscillation amplitudes are
known for all values of n l9 n 2 , 7i 3 , and a. The general solution is equal
to the sum of partial solutions due to the linearity of electrodynamical
equations (27.2):
A (r, =27 A < k > r > 27< A 2 c ' kr + Are"") . (27.5)
k, a k, a
Energy of the field. An electric field is calculated in accordance with
(12.29) and (27.5):
E =  = 27< A 2* kr + AT *'' kr )  (27.6)
k,o
The amplitudes of the field depend harmonically upon time, so that
A = *<o k A, A*=fcco k A. (27.7)
Therefore
E = 4j7 co k( A Z e '' kr  Ajfe'*)  (27.8)
k,<7
The magnetic field is determined from (21.28) and (12.28):
H  rot A =2?([ Ve '' kr > A k1 + H? e ~ l kr > A k*D =
k,0
 iJTakAjp e'' kr  [kA*]e "). (27.9)
k,a
Let us now calculate the field energy. According to (13.21)
dV. (27.10)
To obtain E 2 we perform summation over k, k', cr, and cr':
k.k'a.o'
 A* Afci ( k  k/ )' + A*A*e'( k k/ )') . (27.11)
It is expedient, when integrating jE/ 2 over a volume, to change the
order of summation and integration, each volume integral being
18  0060
274 QUANTUM MECHANICS [Part III
reduced to the product of three integrals of the following type:
 (,+ ') n (r 2ni ("i + n i> }_1
= e*> l l dx^^. t ~~4~ l 0
J 2 7^ (>*! + <)
o
/V71 , X
(27.12)
for n l f T&I = 0.
If Tij + ??i this integral is equal to a x . Therefore the triple integral
assumes one of two values:
f''J;^!!; ,27,3)
It follows that in the expression for E* the double summation with
respect to k and k' becomes a single one after integration, and we must
replace k' by k in the terms involving the product A A''. In
terms containing A Ak'* we replace k' by k due to the factor
c~' k/r . Thus,
V =  7JT ( AJ A^ k  A A?  A* Ajf + A f A^ k ) . (27. 14)
k, c, o'
But (given a / a') A and A^ k , AJ and AJ'* are orthogonal
vectors. Therefore, instead of the double summation with respect to or
and a', there also remains a single sum with respect to <r:
JE*dV=  5J[>k( A k A lk+ Af A^ 2 A A*) (27.15)
k,o
When calculating the integral of the square of the magnetic field
we make use of the rule (27.13). But since the product [k' A']
is replaced by [kA^ k ] if k'  k, we get
J + 2[kA] [kAj'J). (27 J 6)
k, a, o'
The vector products may be expressed by known formulae:
(kAjp (kAf )  FAAf, (27.17)
where we have used the trans versality condition (27.3). This expression
becomes zero for a/ CT'.
Thus,
JH*dV= 7jT& 2 (AAl k +A*Al* k +2AA<,*). (27 . 18)
k,o
Combining (27.14) and (27.18) and taking advantage of the fact
AU~x .2 ^2 /2  __ x
Sec. 27] QUANTIZATION OF THE ELECTROMAGNETIC FIELD 275
o>AA*. (27.19)
k,o
Passing to real variables. In order to apply the usual equations of
quantum mechanics to the electromagnetic field, it is convenient to
pass to real variables
n / 5 /' a  r> CT \
(27.20)
< 27  21 )
where e is a unit vector in the direction of polarization of the wave.
We get $ expressed in terms of a sum of energies, or of the Hamiltonian
functions for linear harmonic oscillators:
 < 27  22 )
If we regard P and Q as ordinary classical dynamical variables,
they satisfy the equations for a linear harmonic oscillator. Indeed, it
follows from (27.22) and (10.18) that P = <4 Qi and Qg = P2.
This agrees with the harmonic time dependence of the amplitudes
Ag, A.
Each separate oscillator is characterized by four integers n v n 2 . n 3 ,
and cr, which label the independent degrees of freedom of the electro
magnetic field. Qk are normal coordinates of the electromagnetic
field [see (7.31)].
Quantization of the electromagnetic field. The result we have ob
tained (27.22) is of fundamental significance. It provides the most
simple and vivid method of applying quantum mechanics to the electro
magnetic field. Indeed, the equations of nonquantized oscillators are
equivalent to the electrodynamical equations of a nonquantized electro
magnetic field, the only difference being that they describe the field
in other variables. But the oscillator problem in quantum mechanics
having already been solved in Sec. 26, quantization can be performed
in these variables as before. Quantization of the motion of oscillators
representing a field in vacuo is just this quantization of the equations
of electrodynamics performed in the appropriate system of variables.
From (26.21), the energy of an oscillator in the Nth quantum state
was
Therefore, with the aid of equation (27.22), the energy of the electro
276 QUANTUM MECHANICS [Part III
g = J^cs.^ (# w ,. + I) 2>" k KO+ Y) ( 27  23 )
n a . <* k, o
The numeral N ni>thitl3ta gives the number of the quantum state of
the oscillator, classified by the numbers n v 7i 2 , 7i 3 , and the polari
zation a.
Quanta. We see that the energy of an oscillator can experience
increments equal to Ao> Ml , M2fM3 . This quantity of energy is called
the energy quantum of the electromagnetic field. Disregarding the
zero energy * < V"*" for the time being, we see that the field energy
is equal to the sum of the energies of its quanta &co nitflfffv Thus
is found a quantum expression for the energy of an electromagnetic
field. It will be shown in exercise 1 of this section that the momentum
of a field is equal to the sum of the momenta of its quanta, while the
momentum of each quantum is found to be related to its energy by the
expression
p = Ak = ^. (27.24)
Thus, a quantum possesses the properties of a particle of zero mass.
The possibility of the existence of such particles was elucidated in
connection with equation (21.16).
Polarization of quanta. A quantum has one more, so .to say, internal
degree of freedom, that of polarization. This peculiar degree of freedom
corresponds to the "coordinate" a, taking only two values <r~l
and a = 2. The energy does not depend upon CT. But, of course, in order
to fully define the oscillation corresponding to a given quantum, we
must indicate the number or = l, 2 as well as the three numbers n l9
n 2 , 7i 3 . We observe that these quanta relate only to the transverse
field of electromagnetic waves and do not describe the Coulomb
field.
The classical approximation ot quanta and electrons. A quantum
should by no means be regarded as the outcome of some mathematical
sleight of hand that led us to equation (27.23). The quantum is an
elementary particle just as real as the electron. For example, when
Xrays are scattered by electrons, the energy of each individual quan
tum Ji to and its momentum &k are involved in the general law of
conservation of energy and momentum in collisions in the same way
as for any other particle. In scattering, the frequency of a quantum
diminishes, in direct proportion to its energy.
The essential difference between the quantum and the electron
consists in the fact that in classical theory there is nothing to which the
quantum corresponds; its energy <?<,>= A to and its momentum p=Ak
tend to zero when h tends to zero. Yet the quantum mechanics of the
electron admits of a classical approximation, since the quantum quan
Sec. 27] QUANTIZATION OF THE ELECTROMAGNETIC FIELD 277
tities for the electron become corresponding classical quantities, which
do not tend to zero even when we take h >0. The relation (27.24)
has a counterpart in classical electrodynamics; the energy density
of an electromagnetic wave was shown in Sec. 17 to be related to its
momentum by means of the factor c. In the limiting transition to
classical theory, the energy of each quantum is regarded as infinitely
small while their number JV kf is infinitely large, so that the wave
amplitude remains finite.
Occupation numbers. Passing from Cartesian coordinates to new
independent variables (the components of the wave vector k), wo
renumbered the radiation degrees of freedom, the quantities Q&
now being the generalized coordinates. The state of a field is specified
if all the "occupation" numbers JVk, o a re known, because the number
N kt a defines the quantum state the given harmonic oscillator is in,
i.e., the number of quanta in the state k, a. The numbers N k , a may
be regarded as the quantum variables of an electromagnetic field.
When a field interacts with a radiator (for example, an atom) these
numbers change. For example, if the number N^ a has increased by
unity this means that a quantum of corresponding frequency, direction,
and polarization has been emitted.
The ground state of an electromagnetic field. Let us now examine
expression (27.23) when N^ ta = 0. In other words, let us determine
the ground state energy of the electromagnetic field. According to
(27.23) it is
MI, 2, W 3 , O
But since the numbers n l9 n 2 , and n 3 run through an infinite set of
values, the sum (27.25) is infinitely large. It must be said that in this
case the theory is not fundamentally defective because the zero
energy itself (27.25), does not appear in any expression ; the field energy
is always measured from the ground state.
At the same time, the ground state ^ of the quantum oscillators
of an electromagnetic field leads to actually observable effects because
the amplitude of a harmonic oscillator in the ground state is not equal
to zero. It takes on all possible values, and, in accordance with (26.22),
the probability of a certain value of Q is proportional to  ^ (Q)  2 .
In an electromagnetic field, the part of the oscillator coordinates is
played by the generalized coordinates Q, in terms of which the field
amplitudes are expressed linearly. Therefore, one can by no means
assert that the field amplitudes are equal to zero in the ground state
of an electromagnetic field (i.e., in the absence of quanta). The prob
abilities of definite amplitude values are given by the harmonic
 >Q a
oscillator wave functions. These functions are equal to e 2h in
coordinates.
278 QUANTUM MECHANICS [Part III
Electronlevel shift produced by the ground state of a field. The ground
state of an electromagnetic field affects observable quantities. One
of the most important effects of this type consists in the following.
Let an electron move in the potential field of a nucleus. The value
of the electromagnetic potentials of a field acting on the electron is
usually chosen as A = 0, 9='.. Only the static Coulomb field has
been taken into account. In actual fact, the potential of the radiating
field must be added to the static potential, for example, in the form
(27.5). As was indicated, this potential must not be considered equal
to zero even when there are no quanta in the field. The field of radiation
affects the energy eigenvalues of the electrons in an atom. They
prove different from what they should be in a purely Coulomb nuclear
field with A = 0, <p = .
The solution of the problem of finding the energy eigenvalues for an
atom, with account taken of the radiation field, encounters considerable
inherent difficulties. First of all, this problem does not lend itself to
a precise solution by means of mathematical analysis. It becomes neces
sary to solve it approximately, taking the correction to the energy
produced by the radiating field as a small quantity. Til actuality, how
ever, a direct calculation of this correction leads to divergent integrals,
i.e., to infinite expressions.
Nevertheless, it is possible to redetermine this correction so that a
finite expression is obtained. To do this, one has to consider the anal
ogous correction for the energy of a "free" electron not influenced by
the external field of the nucleus, and consider the difference of two
infinite integrals. If in doing so we take great care to follow the rela
tivistic invariance of the expressions, the subtraction turns out to be
a completely unique operation and does not contain any indeterminate
quantity of the form oo oo. The final correction does indeed turn
out to be a small quantity compared with the binding energy of the
electron in the atom in its ground state (4 x 10 ~ 6 ev for the hydrogen
atom). The value of the correction is in excellent agreement with mod
ern radiospectroscopic data.
Subtraction of infinities. The meaning of such a subtraction consists
in the following. Physically, an electron is inseparable from its charge,
i.e., from its radiation field. When we talk about a "free" electron,
we always imply that the electron interacts with a radiation field,
which cannot be regarded as equal to zero. The rest energy of an elec
tron is equal to me 2 , where m denotes the observed value of the mass.
In reality, this quantity encompasses the energy of all interactions
of the electron, including interaction with the field of radiation.
Thus, in calculating the energy of an electron in a Coulomb field,
the mass of the electron must be redefined so that, in the absence of
any static external field, all the energy has a finite value we 2 . This
SCC. 27] QUANTIZATION OF THE ELECTROMAGNETIC FIELD 279
redefinition operation or, as it is called, "renormalization" of mass
allows us to find a finite quantity for the energy eigenvalue of an
electron in an atom. Concerning the levels in a hydrogen atom, see
Sees. 31 and 38.
The renormalization operation consists in the fact that the mass
which appears formally in the equations of mechanics, together with
the mass due to the interaction of the electron with radiation, is re
garded as the observed finite quantity w = 0.9 106 x 10~ 27 gm. Only
this known quantity appears in the final result for any calculated
effect.
Difficulties of the theory. The appearance of divergent (i.e., infinite)
expressions in quantum electrodynamics is a defect of the theory and
indicates a certain internal contradiction. The modern form of the
theory as given by J. Schwinger, R. Feynman and F. Dyson, is appar
ently, not yet final.
It is all the more remarkable that despite this imperfection, quan
tum electrodynamics is capable with the aid of renormalization of
yielding correct and unique answers when calculating concrete quan
tities observed in experiment.
Exercise
Calculate the electromagnetic Hold momentum in a vacuum in terms of
the normal field coordinates Q
In the expression (13.27)
we substitute the electric and magnetic fields from (27.8) and (27.9). After
integration over the volume we obtain by (27.13)
k ([A k I kA ' J] + [A [*Ak*JJ + [Af I kA']] + [Af
k, a, a'
We rearrange the double vector products:
" T^Z 1 w * (kA k A  k I 2 kA A* + kA* A'* k )
k, a
The quantities kAk A^_k and kAk* A^!k are odd functions of k, and disappear
after summation over all k. There remains only
k, o
Substituting here the normal coordinates of the field from (27.20) and (27.21),
we arrive at the expression
p E\ 2 <n +4<> 2>* (AV . + I) .
k, o k, o
so that the momentum of each quantum is related to its energy by relation
280 QUANTUM MECHANICS [Part III
Sec. 28. QuasiClassical Approximation
The classical limit of a wave function. It was shown in Sec. 24 that
the limiting transition from quantum mechanics to classical mechan
ics is performed by means of the substitution of (24.5)
= h
(28.1)
By substituting ip into (24.11) (i.e., into the Schrodinger equation),
A
eliminating e h , and formally making h tend to zero, we obtained the
correct classical relation between energy and momentum. This limiting
process signifies that the de Broglie wavelength is very small
compared with the region in which the motion occurs.
It is sometimes useful not to carry the limiting process to its end,
namely in those cases when the asymptotic form of the almost classi
cal wave function is important. If the wavelength is small compared
with the region in which motion occurs, then the wave function has
many nodes in this region and, as we know from Sec. 25, this corre
sponds to an energy eigenvalue with a large number. Thus, the passage
from quantum laws of motion to classical laws is accomplished through
a region in which the number of the eigenvalue is large compared with
unity. If the energy eigenvalue is determined by several integers, as
in the case with motion having several degrees of freedom (cf. the
problem of the potential well), then all these numbers must be large
compared with unity so that the motion should be close to the classical
limit. The wave function in approximation (28.1) (where S is the classi
cally calculated action) thus allows us to determine eigenvalues with
large numbers [see (28.18)].
Solutions with real values of the exponent. The wave function does
not convert to zero also for real values of the exponent in (28.1), i.e.,
for irnaginarjr values of S. Imaginary values can only occur in those
regions of space into which for a given energy the trajectories of
classical motion cannot enter, because the potential energy would be
greater in that case than the total energy, corresponding to a negative
kinetic energy or an imaginary velocity. In these cases, the square of
the modulus of the wave function determines the probability of a
particle penetrating into a classically unattainable region of motion.
Naturally, this probability does not convert to zero only anterior to
the limiting transition, and not posterior to it, when h has already
been eliminated from the equations. This is why approximation (28.1)
is termed quasiclassical.
The quasiclassical approximation. Let us write the equations of
transition to a quasiclassical approximation in a onedimensional
Sec. 28] QUASICLASSICAL APPROXIMATION 281
o Cf /I G
case. From (24.14), by substituting ~^~ = $ and y $= 7 into it
(the onedimensional case!), we have
i7), (28.2)
whence ~) dx. (28.3)
Unlike classical mechanics, equation (28.3) holds not only for
$ > U, when the root is extracted from a positive quantity, but also
when < U, when the action is imaginary.
In Sec. 25 we investigated a similar example of the precise wave equa
tion for the problem of a potential well of finite depth. In the region of
the well , the wave function was of the form fy = sin x x, while outside the
well it approached zero exponentially like e~**. This resulted precise
ly when $ < [7, i.e., in a classically unattainable region. Of course,
an imaginary velocity signifies that a particle moving according to
classical laws does not attain the given position. Therefore, the expres
sion
JV2m"( t/^?)" dx (28.4)
must not longer be understood as action, but simply as the exponent
in the equation
;J s_
fy = e h >
extended to the region <? < C7, if there is such a region. When h ^0,
the wave function will be damped in this region infinitely quickly like
e~ ; and this denotes the unattainability of points where for classical
motion $ < U.
The potential barrier. In the potential well problem, the region for
which U >$ extended rightward to infinity. Therefore, the wave
function became zero at infinity. Considerable
interest is attached to another problem, in
which the potential energy at a certain distance
away from the well again becomes less than
the total energy. This is shown in Fig. 42.
U >$ for the region x l ^ x ^ x 2 . Therefore,
in classical mechanics a particle situated to
the left of x = x l cannot under any circum
stances attain the region x > x 2 , from which x i x z x
it is separated by a potential barrier. In Fig. 42
quantum mechanics, the wave function does
not become zero between x^ and # 2 , since this region is finite (see
exercise 2, Sec. 26).
In the approximation (28.1), the exponent in the equation is a real
quantity when x < x v Therefore, the modulus of fy remains equal to unity :
282 QUANTUM MECHANICS [Part III
On the other hand, between x l and # 2 , the modulus of ip decreases
according to the law
x \2 *
*j dx  4 f VSlToN^n dx
J =e *i . (28.5)
At # = # 2 , in comparison with the point x~x l it diminishes in the ratio
~~ T" V2m(U<f) dx
#=e ' t (28.6)
after which it again stops changing, since S becomes a real quantity.
Hence, the square of the modulus of the wave function diminishes
between x^ and x 2 according to the expression determined by the
quantity B. A more precise theory provides a correction factor for B,
though fundamentally, the function is determined by B alone. B is
called the barrier factor. Somewhat later, it will be explained how the
B factor is related to the penetration probability through the barrier
in unit time.
The quasiclassical approximation is feasible only when the order
of magnitude for the action is large compared with h. This was men
tioned in Sec. 24, when the conditions for the limiting transition to
classical mechanics were being determined. For this reason, equation
(28.6) may be used only when the exponent is large compared with
unity. If it is comparable with unity, the penetration probability
through the barrier must be evaluated by means of precise wave func
tions,
The Mandelshtam analogy. In wave optics there is an analogy to
the passage of a particle through a potential barrier. When a lighl
wave falls on the boundary of a medium of small refractive index froir
a medium of larger refractive index at an angle whose sine is greater
than the ratio of the indices of refraction, there occurs total internal
reflection in accordance with the laws of geometrical optics. If the
problem is solved in strict accord with wave optics, on the basis of
Maxwell's equations (see exercise 2, Sec. 17), then it turns out that the
wave penetrates somewhat into the second medium, but dies out
exponentially in it. L. I. Mandelshtam took notice of this analogy
between quantum mechanics and wave optics. It can be applied in the
following manner.
Let us imagine two optically dense media separated by a layer
optically less dense. Let a light ray fall on the interface of the media
at an angle to the normal larger than the angle of total internal re
flection. According to geometrical optics, the ray should be complete
Sec. 28] QUASI CLASSICAL APPROXIMATION 283
ly reflected by the layer, and it is absolutely immaterial whether or
not there is a denser medium beyond the layer, or whether the re
flection occurs from an infinitely thick, nondense medium. Similarly,
a particle in classical mechanics is completely incapable of penetrating
the barrier. According to the laws of wave optics, light penetrates
into a nondense medium, but dies out in a thickness comparable with
the wavelength., Therefore, if the second dense medium is situated
closely enough, part of the light "seeps" into it.
The classical expression for the amplitude of a light wave may be
regarded as the wave function in relation to a light quantum. The
transition from quantum theory to classical electrodynamics consists
in considering the occupation numbers ^k,o as large (see Sec. 28).
Then the corresponding field amplitudes change to classical ones. For
this reason the Mandelshtam analogy represents an example of quanta
penentrating through a barrier. As has already been pointed out, the
limiting transition for electrons occurs differently; it corresponds to
the transition from wave optics to geometrical optics. Therefore, in
the classical limit, electrons do not penetrate the barrier.
The existence of penetrations through a barrier clearly indicates
that the concept of a trajectory is sometimes completely inapplicable
in the case of quantum motion. A trajectory extended under the
barrier would lead to imaginary velocity values.
Alpha disintegration. Passage through a potential barrier enables us
to explain one of the most important facts of nuclear physics, that of
alpha disintegration. The nuclear masses of heavy elements with atomic
numbers greater than that of lead satisfy an inequality of the form
(21.12):
m (A , Z) > m (A  4, Z  2) + m (4, 2) . (28.7)
Here A is the atomic weight and Z is the nuclear charge (i.e., the
atomic number in the Mendeleyev table). Thus, m (4,2 ) is the mass of
a helium necleus with atomic weight 4 and atomic number 2. Such a
nucleus emitted during alpha disintegration is called an alpha particle.
All that can be seen from equations (21.12) and (28.7) is that the
spontaneous decay of a nucleus of mass m (A, Z) is possible, though no
indication is obtained about the time law of disintegration. The
nuclei of certain elements have mean decay times of 10 10 years while
others have decay times of about 10~ 5 sec, which is a difference of
23 orders of magnitude. It will be noted that the energy of the alpha
particles emitted differs here by a factor of only two. From experiment
it turns out that the logarithm of the mean decay time of a nucleus
is inversely proportional to the alphaparticle velocity. It is this
logarithmic law that corresponds to the difference of 23 magnitudes.
It is accounted for by the difference of barrier factors which depend
exponentially upon the energy.
284 QUANTUM MECHANICS [Part III
The potentialenergy curve. At large distances from a nucleus, an
alpha particle experiences a repulsive force of potential energy
[cf. (3.4)]. At small distances, attractive forces must act because,
otherwise, the nucleus (A, Z) could not exist at all. We do not know
the force law (i. e., the shape of the potentialenergy curve when the
alpha particle is situated sufficiently close
to the nucleus) and, therefore, in Fig. 43
we draw it at will. In this we must be
guided by the following considerations.
The special nuclear forces which hold
the alpha particle in the nucleus before
emission have a smalJ radius of action, so
the potentialenergy curve has the form
of a "potential well." Motion inside the
nucleus corresponds to motion inside such
a well. The transition region from the well
to the Coulomb curve is not very essential
 43 for final results, i. e., it little affects the
exponent of the barrier factor.
The barrier factor for alpha disintegration. The energy of an alpha
particle is positive at an infinite distance from the nucleus. It is
this that signifies that the nucleus is capable of alpha disintegration,
i.e., the alpha particle can move infinitely. In order to find the prob
ability for alpha disintegration, we must calculate the barrier factor
B in accordance with (28.6). Because nuclear forces are shortrange
forces, the transition region is small and we can extrapolate the
Coulomb law, without sensible error in the integral, up to the point
r = r 1 where S becomes greater than U. Point ^ is the effective nuclear
radius determined from alpha disintegration. Other data concerning
the nucleus lead to somewhat different values for the respectively
determined effective radius. This is understandable since r x is obtained
on the particular assumption that the Coulomb law is valid up to the
region for which the potential energy curve is taken in the form of a
well with steep sides.
And so we determine the barrier factor according to the equation
(28.9)
ubsti
2]Z_2)e 2 "" *' (28.10)
The integral in the exponent can be easily calculated by the substi
tution
Sec. 28] QUASICLASSICAL APPROXIMATION 285
Then, after elementary treatment, it reduces to the form
(08 U)
^' j
The quantity r^^ 1 2 is the ratio of the alpha particle energy to
the effective barrier height at the point r l9 taken according to equation
(28.8). Let us evaluate this ratio. For heavy nuclei 2 (Z 2)^180,
r t ^ 9 x 10 13 cm ; we shall take S to be equal to 6 Mev, e 2 = 23 x 10 ~ 20 .
Whence
<?>! 6. l.
__ _
2 (Z  2) e* ~~ 180 23". 1Q 20  ~ 5 '
To a first approximation, we shall consider this quantity as small.
Then, on the righthand side of (28.1 1), we obtain the approximation
2\/2i
n g "/ ^'^ \
2 ' 2(Z~2)e 2 ]~
= . 2 (Z  2)  V^7 2 " ("Z^2)~ . (28.12)
It is easy to check the correctness of this expansion by a direct numer
ical substitution.
The time dependence of alpha disintegration. We shall now show
how the expression for the barrier factor is related to the probability
of alpha disintegration in unit time. In exercise 2, Sec. 25, it was found
that the particle flux passing through a barrier is proportional to the
particle density before the barrier. It is easy to see that the basic
result obtained in the problem of a rectangular barrier coincides,
in the limit, with the result of this section, if we go over to the quasi
classical approximation. Indeed, we, so to say, divide a barrier of
arbitrary shape into separate, successive rectangular barriers. The
penetration probability for each of them is e h =
_ e  2x &x 5 where A a; is the width of the rectangular barrier. The total
penetration probability will be determined by reduction of the ampli
tude of the wave function over the whole width of the barrier. In other
2 _____
words, it will be proportional to the product II e h . This
i *. t, i i u j. j vi ""^7"T Vam(Utf) Ax
product can, obviously, also be represented like e ^' "
286 QUANTUM MECHANICS [Part III
or, in the limit, like e h } m * corresponding to (28.6). Thus,
for any potential barrier, we can assert that the flux of transmitted
particles is proportional to the particle density before the barrier
multiplied by the barrier factor.
From this we can deduce the time dependence of alpha disintegra
tion. The probability of an alpha particle existing inside the nucleus
is equal to the integral of the square of the wavefunction modulus
over the volume of the nucleus, i.e., over the region r<r l . As has
just been indicated, the alphaparticle flux emitted from the nucleus
is proportional to the probability density of their being in the nucleus,
the constant of proportionality being basically determined by the
barrier factor. It follows that the number of nuclei decaying in unit
time is proportional to the total number of nuclei present that have
not disintegrated by the given instant. The constant of proportionality
depends upon the shape of the barrier and upon the state of the par
ticles inside the nucleus, but it cannot be a function of time, as may
be seen, for example, from the equations obtained in exercise 2, Sec. 25.
Indeed, this constant is obtained from the solution of the wave
equation with the time dependence eliminated, i.e., (24.22).
For this reason, the law for alpha disintegration is expressed by the
equation
_ rt
^L = __ JL N 9 N=N Q e '", (28.13)
where N is the number of nuclei that have not disintegrated at the
given instant of time and N Q is the initial number of nuclei. The
quantity F has the dimensions of energy for convenience of comparison
with other quantities of the same dimension. Every nucleus has the
same probability of decaying in unit time, no matter how long it
has been in existence. This probability is = and does not depend
upon time.
Equations (28.13) and (28.12) confirm the experimentally deter
mined law which yields an inversely proportional relationship between
the logarithm of the probability for alpha disintegration and the
alphaparticle velocity v.
The nuclear wave function before decay. It is easy to obtain a
time dependence for the wave function of a nucleus which has not
emitted an alpha particle. It looks like
^elr.e"'^" (28.14)
The first factor accounts for the exponential attenuation of ampli
r<
tude according to a e 2/l law (since the probability, i.e., the square
Sec. 28] QUASICLASSICAL APPROXIMATION 287
Tt
of the amplitude, diminishes according to e h ) ; the second factor
is the usual wavefunction time factor. Expression (28.14) is very
similar to the wellknown formula for damped oscillations, with the
difference that in the given case it is the probability amplitude
of the initial (not yet decayed) state of the nucleus that is
damped.
The wave function (28.14) satisfies the initial condition <p (0) = 1.
Since the wave function equation is of the first order with respect
to time, it is absolutely necessary to impose some initial condition.
It was assumed that at the initial instant of time an alpha particle
was definitely situated inside the nucleus. However, for this wave
function, the righthand side of equation (24.16) does not become
zero : there is a finite flux for the probability of a particle being emitted
from the nucleus. This is what leads to the exponential damping
of the probability of the predecay state.
All nuclei before disintegration are described by exactly the same
wave function (28.14), if at the instant = they were in the initial
state. Therefore, they all have a perfectly identical probability of
decaying in unit time, and it is impossible to predict which one of
them will decay earlier and which later. In exactly the same way,
in the diffraction experiment it is impossible to say which part of
the photographic plate will be hit by a given electron. The decay
law is purely statistical, in the same way as the law for diffraction
pattern.
In this sense, radioactive decay does not resemble the falling of
ripe fruit from a tree an alpha particle in a nucleus is always identi
cally "ripe" for emission. This is suggested by the homogeneity of
the decay law (28.13) with respect to time.
The indeterminacy of a nuclear energy level before decay. The wave
function (28.14) does not belong to any eigenvalue of the energy S,
since the states with energy $ have wave functions which are time
i^~
dependent according to e h . Such states should exist for an un
limitedly long time because the amplitude of their probability does
not fall off, while the probability for an undecayed nuclear state
is reduced by e times in a time A == . We can define the time interval
* i
as the characteristic (or mean) attenuation time.
Let us now suppose that the wave function (28.14) is represented
as the sum of wave functions of states whose energies are determined
accurately. We know that the time dependence of these functions
is given by the factor e h . In other words, we have to represent
__ _EL_ . <*<>'
the * 'nonmonochromatic" wave e 2h h as the sum of "mono
288 QUANTUM MECHANICS [Part III
chromatic" waves f ,"~ l 'ir*\ I* 1 which energy interval A<? (by order
of magnitude) will the amplitude of these monochromatic waves
differ noticeably from zero ?
We can answer this question by making use of relation (18.6).
According to this relationship, a wave of duration A is represented
by a group of monochromatic waves whose frequencies lie in the inter
val A <o >  ~ ~ . Substituting, in place of Aco, the equivalent
<"'' t\t fi \ JP
(in this case) quantity ' , we arrive at the following estimate:
Af A*> 27C&. (28.15)
This is the uncertainty relation for energy. The measure of
uncertainty for the energy is the quantity P, i.e., the inverse value
for the probability of decay in unit time. (28.15) should be formulated
thus: an energy of a state existing during a limited time interval A
is determined within the accuracy of the order of TT  Only the
energy of a state which exists an uiilimitedly long time is fully deter
minate.
The meaning of the uncertainty relation for energy. The meaning
of the uncertainty relation for the coordinate and the momentum
(23.4) is not analogous to the meaning of (28.15). The estimate (23.4)
expresses the fact that the coordinate and momentum do not exist
in the same state; (28.15) signifies that if a state of the system has
a finite duration A, then its energy at each instant of time within
the interval A is not determined exactly, but is only contained
within a region of the order of P.
The quantity P is termed the level width of the system. The concept
of level width can be applied to any states of finite duration and not
only to alpha disintegration. For example, the energy level of an
atom in an excited state has a definite width, since an excited atom
is capable of the spontaneous emission of a quantum.
Explanation of the level width. We shall now show how the level
width of a nucleus capable of alpha disintegration can be found
by considering the wave function variation under a potential barrier.
In was shown in Sec. 25 that infinite motion has a continuous
spectrum. The motion of a system with a potential barrier is infinite
because an alpha particle is capable of going to infinity. It follows,
strictly speaking, that a nucleus capable of alpha disintegration
should have a continuous spectrum.
* A wave with definite frequency is termed "monochromatic." A mono
chromatic wave corresponds to a single colour (chromos is the Greek for colour).
Ser. 28] QUASICLASSICAL APPROXIMATION 289
Let us now evaluate the energy level widths F for alpharadioactive
nuclei. From (28.15) it follows that even for a nucleus with a very
short alphadecay time (~10 5 sec) F~10 22 erg~0.6 x 10 10 ev.
How is it possible to combine a continuous spectrum with such a
narrow energy interval ?
The solution to the wave equation between r t and r 2 is of the
following form:,
T
y f V 2m(U#) dr j I V 2m(U <f) dr
^ = C,e ' ( + 2 e l ;[ . (28.16)
The first solution exponentially diminishes with r, while the second
exponentially increases. It follows that if the barrier extended to
infinity rightwards, a solution would exist only for G 2 = 0. The ratio
C 2 /C ly determined from the boundary conditions at r = r v is a function
of energy. It is the roots of equation C 2 (} = that give the possible
energy eigenvalues for finite motion. The energy of a particle in a
well of finite depth is obtained in just this way. A particle in a well,
which was considered in Sec. 25, differs from a particle beyond the
barrier in that the barrier is of finite width. Therefore, the second
solution, proportional to C 2 , need not be strictly equal zero, but
may only be small compared with the first solution in any small
interval of values $ close to a root of equation 2 ($} = 0. This region
of values of & is what corresponds to the assumption that the modulus
of the wave function outside the nucleus is small compared with the
wave function inside the nucleus. In other words, if the energy of
the nucleus is contained in a given region of values F, then we can
say that the alpha particle is in some way bound in the nucleus,
similar to the way that a particle can be bound in a real potential
well. The higher or broader the potential barrier, the less the barrier
p
factor B and the less the decay probability T proportional 'to it.
But then &$ is also correspondingly reduced, i.e., the continuous
spectrum state becomes closer to the discrete spectrum state with
an exact energy value < . This is what explains the meaning of
the uncertainty [$\ it indicates how close the state is to a bound
one, with an infinitely long lifetime.
The uncertainty of &# does not limit the applicability of the law
of conservation of energy in any way; the total energy of a nucleus
and alpha particle is constant. However, the state with a strictly
defined energy relates simultaneously to disintegrated and non
disintegrated nuclei, while a nondisintegrated nucleus has an inexactly
defined energy.
Any state capable of a spontaneous transition to another state
with the same energy possesses a certain energy width. The energy
290 QUANTUM MECHANICS [Part III
is not defined exactly in each of these states separately, but a precisely
defined energy corrasponds to both states at once.
We can divide the total level width into partial widths related
to the probabilities for various transitions. Thus, strongly excited
nuclear states are capable of emitting neutrons of various energies
and of radiating gamma quanta. Each possibility contributes its
exponential in the term characterizing attenuation. The total atten
uation is determined by the product of such exponentials. It follows
that the total level width is equal to the sum of its widths in relation
to all possibilities of disintegration.
The Bohr quantum conditions. Let us now apply the quasiclassical
approximation to the finite motion of a particle in a potential well
and find the energy levels. From (28.3) the wave function is
X
fy  sin (~ J dx V 2m (tf U)' + y) . (28.1 7)
* i
The real solution involving the sine is taken, since, in accordance
with (24.20), it does not involve any particle flux outside the well,
i.e., it corresponds to a stationary state.
Here x 1 is the left edge of the well for which $ = U. In changing
x to x l to the righthand edge x 2 , the phase of the wave function
can change by a whole multiple of TT together with a certain addition,
which we shall call (3 for the time being. In the whole length of the
well, the sine changes its sign for a given value of the particle energy
by as many times as TT is added to the argument of the sine, i.e.,
the wave function (28.17) has as many nodes. However, the number
of nodes is equal to the number of the energy eigenvalue n ; therefore,
an equality is obtained for the determination of <& n :
**
i J V 2m (ff n  U) dx = KH + p. (28.18)
*i
If the edges of the well are not plumb at the points x l and x 2 ,
then a very fine analysis shows that P = v It turns out that (3  =
for plumb edges of the well, because the condition fy (a^) fy (x 2 ) =
is then imposed on the wave function, as in the problem of an infinitely
deep rectangular well. In addition, we note that the points x l and
#o are given by the result that & U (xj^ U (x 2 ), i.e., x and x 2
depend upon S.
It should be noted that, from its very meaning, equation (28.18)
holds only for a large n, to a quasiclassical approximation. The integral
is considerably greater than h for large n, and this signifies that
$ (# 2 ) S(x l )^>h, whence, to a quasiclassical approximation, we
obtain (28.17).
Sec. 29] OPERATORS IN QUANTUM MECHANICS 291
Equation (28.17) was postulated by Bohr in 1913 in determining
the stationary orbits for a hydrogen atom (P was considered to be
equal to zero). Bohr supposed that electrons in such orbits do not
radiate light and do not falJ onto the nucleus, while radiation occurs
only in the case of a transition from one orbit to another.
Thus, it turns out that the Bohr quantum conditions emerge as
a limiting case, of quantum mechanics without any additional
postulates.
Exercises
1) Determine the energy levels for a linear harmonic oscillator from equation
(28.18).
From this ^ = /&coln+ I which is fortunately correct for all n and not only
for n ^> 1.
2) Find the approximation that follows (28.3).
We look for S in the form S = S + hS l . Then
d^
dx*
The zero approximation gives S from (28.3). The first approximation yields
"" or
so that ib = ,____ e
3) Find the factor B for a barrier of the form U = when x < 0, U *= U oa?
when x>0, # < U .
Sec. 29. Operators in Quantum Mechanics
Momentum eigenvalues. In a number of cases we are able to deter
mine energy eigenvalues from the wave equation (24.22). However,
it is very important to find the eigenvalues of other quantities too :
linear momentum, angular momentum, etc. To do this, it is con
venient to proceed from the form of ^ in the limiting transition to
classical mechanics:
,A
9 = e h . (29.1)
292 QUANTUM MECHANICS [Part III
7 r\
Let us apply the operation 3 to both sides of equation (29.1),
1i G X j
i.e., we take the partial derivative with respect to x and multiply by 4 :
But in the classical limit S becomes the action of the particle, while
o Of
 becomes the component of momentum p x [see (22.9)]. Therefore,
ox
the equation for the momentum eigenvalues that yields the correct
transition to classical mechanics is of the following form:
where p x is the eigenvalue for the xth momentum projection.
Momentum and energy operators. Let us compare equation (29.3)
with the wave equation (24.22):
Here, the symbol I . ) ty denotes ~ and similarly for ( ^ 1 fy ,
(0 \2 * \^ x i u x \vy/
a) *
In order to find the energy and momentum eigenvalues we must
perform a definite set of differential operations and multiplications
by the function of coordinates in the left part of the equation. But
these sets are connected in a very curious manner, as will now be
shown. We shall call the symbol = , multiplied by . , the momentum
* /? ft
operator applied to a wave function. Instead of y we will
symbolically write p x . Then, it will be necessary to rewrite equation
(29.3) as
M = p x ty . (29.5)
This equation denotes exactly the same as (29.3), though the sym
bolic notation p A should emphasize that the corresponding operation
is applied in order to find the momentum eigenvalues.
The operation on the lefthand side of (29.4) we shall also symbol
ically call^. We write <& and not & because the energy is assumed to
be expressed in terms of momentum, similar to the Hamiltonian func
tion #*. Then, in shorter notation, (29.4) appears as
<!f4/. (29.6)
is called the Hamiltonian operator, or the energy operator.
Sec. 29] OPERATORS IN QUANTUM MECHANICS 293
Comparing (29.4) and (29.3), we see that the momentum and energy
operators are related by the same equations as the corresponding
quantities :
*) + U (29.7)
We have written U instead of simply U in order to emphasize that
in this equation the expression U is not regarded as an independent
quantity but, instead, as an operator acting upon <p, i.e., a multipli
cation operator of fy by U. Equation (29.7) is symbolic. It is under
stood that both sides are applied to fy.
The meaning of operator symbolism. The usefulness of an abbreviat
ed operator notation in quantum mechanics consists in the fact that
the equations thus become more expressive. The relation between
quantum laws of motion and classical laws, which are limiting cases
with respect to the quantum ones, can be best of all seen in operator
notation.
If in classical equations relating mechanical quantities we replace the
momenta by their operators, we then obtain correct operator rela
tionships of quantum mechanics. The limiting transition to classical
mechanics restores the usual relationships between quantities. Indeed,
in the limiting transition (29.1), the operator p= y w ^ gi ye P^
If we must perform the limiting transition for p 2 , then we need to
differentiate only the exponential each time, because this yields the
quantum of action in the denominator. For h ^0, only terms with the
highest degree of h in the denominator remain, and it is these very
terms which are obtained in replacing the operator p by the quantity
VS (i.e., by the classical momentum vector). We had an example of
such a transition in Sec. 24 [see equations (24.13) and (24.14)].
The angularmomentum operator. It is now easy to define the angu
larmomentum operator. We shall begin with one component M z . It
is clear from Sec. 5 that the angular momentum M z is at the same time
a generalized momentum corresponding to the angle of rotation about
the zaxis, i.e., M z = p cf) . Then, from (10.23)
* < 29  8 >
Therefore, in quantum mechanics the operator p v must be of the form
*.T < 29  9 >
At the same time, in accordance with classical mechanics, the pro
jection M z is related to the momentum projections thus:
M z = xp y ~yp x . (29.10)
294 QUANTUM MECHANICS [Part III
It follows that there must exist an operator relationship
*p y yp*. (29.11)
Let us check to see that the definitions (29.9) and (29.11) do, indeed,
coincide. Let us pass to cylindrical coordinates:
(29.12)
(29.13)
From this we have
dx dr dx $9 dx
Solving (29.12) and (29.13) with respect to x and y, we have
r A/ir 2 + y 2 , 9 arc tan ~ ;
i/ 1
=coscp, g^y/", 8 " 1 '*';
' + 2/ a
y sin 9
Substituting all these expressions into equalities (29.14) and (29.15),
and substituting the derivatives themselves into (29.11), we can see
that both definitions for M z (29.9) and (29.11) are identical.
Angularmomentum projection eigenvalues. Let us now find the
eigenvalues for M z . For this, it is necessary to solve the equation
M z fy = M z fy , (29.16)
i.e.,
IL'^L^MZ^J. (29.1?)
1 C/9
This equation is very simply integrated:
M
^e'Jr ( 29  18 )
As we know, a wave function is the probability amplitude. Function
(29.18) is the amplitude of the probability that the particle possesses
an azimuth angle 9, if the 2th component of its angular momentum
is equal to M z . It is essential that not only the absolute value of the
wave function has physical significance, but also its phase; this is
indicated, for example, by the phenomenon of electron diffraction.
For the phase of the wave function (29.18) to be determined, it must
either not change at all or only by a multiple of 2 n in rotating the
Sec. 29] OPERATORS IN QUANTUM MECHANICS 295
coordinate system through 360; this is because the position of the
particle relative to such a system rotated through 360 does not
change. If in this case the wave function were not returned to its
initial value, it could not uniquely represent the probability ampli
tude. Thus,
i.e.,
e' V^e'^, < 29 ' 19 >
so that
a 1 '"* _!_,** (29.20)
C.  A  ( ,
where k is an integer of any sign or zero. Whence we obtain the eigen
values of M x \
M z =hk. (29.21)
M z is called the orbital angularmomentum projection for a particle.
We shall see in Sec. 32 that a particle can have an angular momentum
connected with its internal motion, which is not described by the wave
function (29.18). Here, we have proven that the orbital angular
momentum components can only assume values which are whole
multiples of h.
The SternGerlach experiment. The discreteness of the angular
momentum spectrum is confirmed by direct experiment. The idea
of the experiment consists in the following: a direct relationship
exists between the orbital angularmomentum projection and the
magnetic moment projection [see (15.25)]:
A narrow beam of vapour of the substance under investigation
is passed between the poles of an electromagnet in a strongly inhomo
geneous field ; to achieve this, one of the poles may be made tapered.
The particles in the SternGerlach experiment, they are atoms
enter the field parallel to the edge of the taper, i.e., they move in
a direction perpendicular to the plane of the lines of force of the
field. The plane of symmetry of the field passes through the edge
of the taper and the initial direction of motion of the particle. We
assume the zaxis to be perpendicular to the edge of the taper and to
lie in the plane of symmetry of the field. If the mechanical moment
of the electrons in the atoms has only discrete, integral projections
on the zaxis, then the magnetic moment of the atoms is established
in several definite ways. The deflecting force acting on the magnetic
moment in a magnetic field is, by (15.40)
(29.23)
296 QUANTUM MECHANICS [Part III
In the plane of symmetry of the field, H is directed along the zdirection
and depends only upon z.
Since the angular momentum can only have a definite set of values,
the deflecting force acting upon the atoms in the beam is also not
arbitrary but has a very definite value for particles with respective
angularmomentum projection M z =hk. It can be seen from (29.23)
that the force is a quantity which is a multiple of 5 ~ Therefore,
the particles in the beam experience only those deflections in the
magnetic field which correspond to the possible values of the force
(29.23). In other words, the beam is split into several separate beams
and does not proceed continuously, as would be the case for any
nonintegral projections M z .
Where the beam is formed, each particle was given a certain angular
momentum. Motion in the magnetic field makes it possible to measure
the projection of this angular momentum M z in the direction of the field.
The impossibility of the simultaneous existence of two angular
momentum projections. From the fact that the angularmomentum
projection on any axis is integral, it follows that the angular mo
mentum does not have, simultaneously, projections on two axes
in space.
Indeed, in the SternGerlach experiment, the zaxis is absolutely
arbitrary. We could have measured the angularmomentum projection
011 some axis in space and then pass the same beams through a
magnetic field making a very small angle with the field in which
the first measurement was performed. Both measurements will givo
only integral projections of the angular momentum. Both one and
the same vector cannot simultaneously have integral projections
on infinitely close, but otherwise arbitrary, directions; when tho
first measurement was performed, the angular momentum had a
projection only on the first direction of the field, and, correspondingly,
in the second measurement, it had projections only on the second
direction of the field.
Similar to the way that coordinate and momentum do not exist
simultaneously, it turns out that two angularmomentum projections
do not exist in the same state.
The simultaneous existence of two physical quantities. We shall
consider from a general point of view the question of which quantities
of quantum mechanics can exist in the same state of a system. Let
us suppose that in a certain state described by the wave function A
there simultaneously exist two physical quantities X and v. This
means that the wave function fy is an eigenfunction of the two operators
X and v. It satisfies two equations
^ = \fy (29.24)
and vij;=v^. (29.25)
Sec. 29] OPERATORS IN QUANTUM MECHANICS 297
X and v are, speaking generally, differential operators; X and v are
numbers.
Let us apply the operator v to (29.24). Since there is a number X
on the right, it can be put on the left of the operator sign v :
vX^ = vXtJ> = Xv^ = Xv<{> . (29.26a)
In the last equation we made use of (29.25). We shall now apply
the operator X to (29.25):
Xv^ = Xv<J> = vX^ = vX^ 1 . (29.26b)
Let us subtract (29.26b) from (29.26a):
vX^ Xv^ = XviJ; vX^ = . (29.27)
(29.27) can be symbolically written as an equality between operations.
5X = Xv or vX Xv . (29.28)
This symbolic equality means that the result of operating with v
and X should not depend upon the order of their actions, otherwise
equations (29.24) and (29.25) cannot have a general solution.
We can also prove the inverse theorem : if two operators are com
mutative, i.e., the result of their action does not depend on their
order, then it is possible to form a common eigenfunction ty satisfying
equations (29.24) and (29.25).
Commutations of certain operators. Let us now apply the obtained
result to two quantities which definitely do not exist in the same
state: the coordinate x and momentum p x . We must calculate the
commutation p x x xp x .
Changing from symbolic notation to the usual one, we obtain
h 8 1 // d$ h j h dty h dfy h />o .>Q\
~~* 5 X T "~~" X ' o ~ w ~T~ X ; ~ ~"~ X ~~i r,  r* w . \ ) , ~j ** )
^ dx T ^ dx * T ^ dx i dx i T v '
Reverting to symbolic notation, we represent (29.29) in the following
form : h
p x x xp x = ~. (29.30)
Thus, the result of operating with p x and x depends upon the order
of their action; p x and x are noncommutative. And this was to be
expected because the quantities x and p x do not exist simultaneously.
The eigenfunction of the operator x satisfies the equation (x x')fy =0.
Consequently, it is equal to zero over the whole region where the
coordinate x is not equal to the chosen eigenvalue x'. This function
differs from zero only at one point x=x'. The eigenfunction of the
momentum operator which satisfies equation (29.30) is e* h ' it
differs from zero over all the space. This example shows how great
is the difference between the eigenfunctions of noncommuting operators.
298 QUANTUM MECHANICS [Part III
The abreviated notation of (29.30) is a convenient representation
of (29.29). In the more complex cases that will be examined later
the convenience of this abbreviated notation is obvious. It must
be borne in mind that the operator notation is simply a rationalization
of mathematical symbolism, and there is nothing incomprehensible
in the result that p x x x p x ^Q, such is the property of operator
symbols. It will be recalled that in vector algebra there also exists
noncommutative multiplication; and, what is more, not of symbols,
but of quantities. In quantum mechanics the operator symbolism
is most expedient.
The various momentum components are commutative:
p x p y p y p* = (29.31)
(the word "operators," will be frequently omitted in future as being
selfevident). The commutative relation (29.31) is obtained simply
from the fact that the result of applying two partial derivatives
does not depend upon the order of differentiation.
It is also obvious that
p y x  xp Y = 0. (29.32)
We now calculate the commutation of any two angularmomentum
components. Let us take M x and M y :
My = zp x xp z .
Let us first of all wite the commutation without using the rules
(29.30)(29.32):
M x M Y MyM x =(yp z zp y ) (zp x xp z ) (zp x xp
We now group the terms here so that the order of coordinates and
corresponding momenta is not disturbed:
M x M y MyM x ^ yp x (pzz zp z ) xp y (p z z zp z ) .
We substitiite the commutation relation p x z zp z i and then find
the required result:
M x M y  M y M x  ih (xp y  yp x ) ^ihM z . (29.33)
Now changing the indices x, y, z cyclically, we obtain the remaining
commutation relations :
My M z M x My = ih M x , (29.34)
M z M x M x M z = ihMy. (29.35^
All three commutation relations can be easily remembered if we
write them in contracted form thus:
Sec. 29] OPERATORS IN QUANTUM MECHANICS 299
[MM] = iAM. (29.36)
Expanding this equality in components, we once again arrive at
(29.33)(29.36).
It will be noted that the vector product of an operator by itself
cannot equal zero unless the operator components of the vector
are noncommutating (but, for example, [pp] = 0).
We have shown that there does not exist a state in which a system
would possess two angularmomentum projections. Angular mo
mentum has a projection only on one axis, in agreement with the
SternGerlach experiment. The only exception is when all three
angularmomentum projections are equal to zero. The eigenfunction
of such a state does not depend upon any angles at all [see (29.18)].
Therefore, as a result of applying differential operations of the type
(29.9), where the differentiation is performed with respect to the
angle of rotation about any arbitrary axis, this eigenfunction is
multiplied by zero. The action of the operators of the angularmo
mentum components on such a function is commutative. This does
not contradict the SternGerlach experiment, because a vector can
have a zero projection on two infinitely close, though arbitrarily
orientated, axes, provided the vector itself is equal to zero. But if
only one of the angularmomentum projections is not equal to zero,
then the two others do not have definite values because, otherwise,
there would be a contradiction in equality (29.36).
The square of the angular momentum. Let us now examine further
properties of angular momentum. We shall show that even though
two angularmomentum projections do not exist, a single angular
momentum projection exists together with its square
M* = Ml + M* + M I . (29.37)
We shall verify this:
because M z and Ml are, of course, commutative. Let us add to the
righthand side of the last equality, and subtract from it the combi
nations M X M Z M X and M y M z M y ; we take M z and M y outside the brackets,
once on the right and another time on the left. Then we obtain
= Q. (29.38)
Here we have made use of the commutation rules (29.33)(29.35).
The eigenvalue of M 2 will be found in the following section.
300
QUANTUM MECHANICS
Exercises
[Part III
1) Find the commutations, of
M X9 px', M x , p y ; M x , p z ;
M x ,x, MX , y ; M Zf z;
M*, p x ; M z , p*, M 2 , x; M 2 , r* .
2) Write down the Cartesian projections of momentum in spherical co
ordinates.
Spherical coordinates are expressed in terms of Cartesian coordinates in
the following manner [see (3.5), (3,7), (3.8)]:
ft = arc cos 
Whence we obtain the partial derivatives
&r x
 = = sin ft cos 9 ,
OX T
dr y
~ ~ = sin $ sin 9 ,
dy r
or z
 = = cos ft ,
GZ r
d 9
dx
3 9
"
dx
80
cos cos
cos 8 sin 9
sin 9
r sin ^ '
cos 9
r sin if '
Further,
.
sm * cos "
coscpcos $ e?
sin 9 a
_a_ __ dr a
cos 9 a
a& a , a 9 a
  =
a 9 '
a ar a
sin 9 cos o a
a a sinft
 = cos ft
3) Write the angular momentum projections on Cartesian axes in terms
of spherical coordinates.
Sec. 30] EXPANSIONS INTO WAVE FUNCTIONS 301
M* = ypzzpy *= jlBin <? cot $ cos 9 ^\ ,
My = zp x xp z = T (cos 9 gy cot sin 9 J ,
iTr ^ ^ **. * ho
M x *p t Vp, rJ j:
4) Write the expression for the square of the angular momentum in spherical
coordinates :
^^
y)
From exorcise 3 we have
Applying M x M*Afy to MX iMy, wo must observe the order of the "factors"
o o
and cv, cot ^ and^ . We obtain
6*9 db
(M x f tJ M y ) (M x  1 M y ) =
COt # 7;
\tf# a c?9
Finally,
1 O
This expression is obviously commutative with M z = 5 . This was
1 o 9
shown in the present section in another way.
Sec. 30. Expansions into Wave Functions
The superposition principle. One of the most fundamental ideas
of quantum mechanics consists in the fact that its equations are linear
with respect to the wave function ^. This result proceeds from the
whole set of facts that confirm the correctness of quantum mechanics,
in the same way as an analogous result in classical electrodynamics
(see Sec. 21), which is also a generalization of experience.
For example, the diffraction of electrons shows that the amplitudes
of wave functions are combined in the same simple way as the ampli
tudes of waves in optics ; diffraction maxima and minima are situated
at the same positions, which are determined only by the phase relation
302 QUANTUM MECHANICS [Part III
ships, independently of the wave intensities. All this points to the
linearity of wave equations; the solutions of nonlinear equations
behave in an entirely different manner.
The sum of two solutions of a linear equation again satisfies the same
equation. It follows from this that any solution of a wave equation
can be represented in the form of a certain set of standard solutions,
similar to the way that, hi Sec. 18, a travelling nonperiodic wave
was represented by a set of travelling harmonic waves (18.1).
The statement concerning the possibility of representing a single
wave function in terms of the sum of other wave functions is called
the superposition principle.
The Hermitian property of operators. Wave functions are usually
represented with the aid of the sum of eigenfunctions of certain
quantummechanical operators. In the present section it will be
shown how such expansions are performed. First of all, however,
it is necessary to establish certain general properties of the operators
whose eigenvalues are physical quantities.
Obviously, these eigenvalues must be real numbers, although the
operators themselves may depend explicitly upon i=V 1 [see
(29.3), (29.10)]. We shall consider the equations for the eigenfunctions
of the operator X and another equation involving its conjugate:
X^X<];, (30. la)
X*^*X*^*. (30.1b)
We must find the condition for which the eigenvalues of the operator
are real numbers, X*=X.
To do this, we multiply (30. la) by ^* and (30. Ib) by ^, integrate
over the whole range of the variables x (upon which the operator X
depends), and subtract one from the other. Then we obtain
4/X* <;*) dx = (X  X*) ft* fy dx .
But the integral of ^*^=^ 2 cannot be equal to zero, since ^ 2
is an essentially positive quantity.
The eigenvalue X is, by definition, real, i.e., X = X*; therefore we
arrive at the relation
* \ty  <; X* <[,*) dx = . (30.2)
Equation (30.2) can be regarded as a condition imposed upon the
operator X. In fact, however, we must demand that the operator X
satisfy the equation (30.2) not only for its own eigenfunctions fy* (X, x)
and $ (X, #), but also for any pair of functions #* (x) and ^ (x), pro
vided these functions satisfy the same conditions of being finite,
continuous, and singlevalued as the eigenfunctions <]> (X, x) :
SeC. 30] EXPANSIONS INTO WAVE FUNCTIONS 303
J(X* ty ~ * ** X*) dx  . (30.3)
The necessity of such a condition will be explained later in this
section. An operator for which equality (30.3) is satisfied is termed
In equation (30.3), dx is an abbreviated notation for dV=dx dy dz,
if the integration is performed over a volume (X ^ J$) y or ^9, if X = M z >
etc.
The Hermitian nature of the operators p x , M z , ... is easily verified
by integrating by parts. For example,
"
00
The eigeiifunctions of the operator M z must satisfy the requirement
of uniqueness (29.19); hence #* (0)=#* (2w), <J; (0)=<; (27e) similar
to the eigenf unctions of the operator .M*. Therefore, the integrated
quantity becomes zero. The operator    is M*, so that
\%
27T
in accordance with the general requirement (30.3).
The Hermitian nature of Jif and M 2 is proven by a double integration
by parts.
The orthogonality of eigenfunctions. An important property of
eigenf unctions follows from the Hermitian nature of operators. Let
us consider the equations for two eigenvalues of the same operator X:
Xip (X, x)  X<J> (X, x) . (30.4a)
X* ty* (X', x)  X' ^* (X', x) . (30.4b)
We multiply (30.4 a) by ^* (X', x) and (30.4b) by fy (X, x), integrate
with respect to dx, and subtract one from the other:
= (X  X 7 ) JV (X 7 , x) ty (X, x) dx . (30.5)
The lefthand side of this equation becomes zero in accordance with
general requirements for Hermitian form (30.3). Therefore, if X'^X,
the following integral must equal zero
(X', x) fy (X, x) dx  for X  X'. (30.6)
304 QUANTUM MECHANICS [Part III
This property was proved in exercise 1, Sec. 24, as applied to the
eigenf unctions of the energy operator. It is called the property of
orthogonality.
Several quantities, X, v, etc., may sometimes exist in the same state.
For this it is necessary that the operators X, v,. . . should be com
mutative. For example, for free motion there exist p x , p Y , and p z .
Then we may form fonctions, which are eigenfunctions with respect
to all the operators simultaneously:
the orthogonality condition for such functions is directly generalized
in the form
J^*(X',v';a;)^(X,v;a:)da; = 0, (30.7)
if XV X or vV v.
Expansion in eigenfunctions. Let us suppose that the eigenfunctions
of a certain operator X are known. These functions always satisfy
(in addition to the equation X^=X^) certain requirements: they are
finite, continuous, singlevalued, and so forth. Then, in accordance
with the superposition principle, any function <L (x) which satisfies
the same requirements may be represented as the sum of the eigen
functions of the operator X:
*(s)=2>'<H x '>*> (30.8)
y
We shall show how to determine the expansion coefficients c\.
To do this we multiply both sides of the equation by ^* (X, %) and
integrate with respect to dx:
(X, x) ty (x) dx =  ft, *) * (*', a) dx . (30.9)
In accordance with the orthogonality condition all the integrals
on the righthand side of the equality (30.9) become zero except those
for which X'=X. Consequently, there remains the equation
,x)\*dx. (30.10)
We shall consider that the eigenfunctions fy (X, x) are normalized to
unity, i.e., j <J;  2 da; = l [see (24.18)]. Then the expansion coeffi
cient is
c x = JV (X, x) fy (x) dx . (30.11)
Sec. 30] EXPANSIONS INTO WAVE FUNCTIONS 305
s*.
In the case when we have a system of commutative operators X,
v, equation (30.11) is directly generalized:
(X,v;*)4;(30d*. (30.12)
The meaning pf the expansion coefficients. We have seen that a
state <{/ (x) is represented as a superposition of states with definite
values of the quantity X. The component of the wave function fy
which corresponds to this value of X is
It represents the probability amplitude for the given value of the
quantity X in the state fy (x). In order to find the probability itself,
w^ 9 of the occurrence of quantity X, we must eliminate the coordinate
dependence, since X and x do not exist in the same state.
To do this, let us integrate the probability density of the state with
a given X, i.e.,  c\  2  ^ (X, x) 2 , over all x. From the normalization
condition for eigenfunctions we obtain
, *) d* =  'x i 2  (30.13)
The quantities Wx = I c x  2 have a basic property of probability:
their sum is equal to unity, provided the function itself satisfies the
condition of normalization (24.18). Indeed,
1 = Jl * (*) \*d* = J 1 2>x * (* *) \*dx =
x
, x)  2 Ax +ZZ c * c *' ** < X/ > *> * < X > *> dx
x x xvx
But on the orthogonality condition (30.6) a double summation is
equal to zero. From this, in accordance with (24.18), it follows that
Ji>x 2 =Ji>x = l. (30.14)
Thus, the coefficient Cx should be regarded as the probability ampli
tude, similar to ty (x). But  fy (x)  2 is the probability of detecting a
particle with coordinate x independently of X, while cx 2 is the prob
ability of finding it with a given value of the quantity X independently
of x.
Expansion in angularmomentum projection eigenfunctions. The
atomic beam in the SternGerlach experiment is split into a certain
number of separate beams, corresponding to the number of angular
momentum components along the magnetic field direction M x =hk.
200060
306 QUANTUM MECHANICS [Part III
Let us denote the largest eigenvalue quantity k by the letter I. Then
it is obvious that
Z >>, (30.15)
i.e., k takes on 21 + 1 values.
The eigenfunction corresponding to M z hk is
67T
(the factor  is introduced for normalization, f tp \*d<p 1) .
V 2rr J
o
If each of the separate beams is once again passed through a magnet
ic field parallel to the zaxis, there is no further splitting; this is
because M z in these beams has a single definite value and not the
whole set of values in the range hi > M z > hi, as was the case in
the initial beam. From this the meaning of the orthogonality of
eigenfunctions is very well seen. If a particle is found in a beam corre
sponding to a given value of fc, then the probability of finding it in a
beam with a different value of the projection M z = hk'^hk is equal
to zero. From the general rule, the probability is equal to the square
of the modulus of the coefficient of the expansion Ck of the function
4 (k) in terms of the functions ^ (&'), i.e., in accordance with the gen
eral expression (30.11)
* 'i
= J
From the orthogonality condition (30.6), the integral is naturally
equal to zero if lc' ^ k. Therefore the orthogonality condition is a neces
sary condition of particles being found in states with definite values
of M z or, as in the case of an arbitrary operator X, in states with defi
nite values of X. But the orthogonality condition follows directly from
the Hermitian nature of operators (30.3), while equation (30.2),
concerning the functions with equal values of X, is insufficient.
The Hermitian condition implies the reality of eigenvalues together
with the possibility of "pure states," i.e., states with definite eigen
values of quantities.
If the second magnetic field is along the #axis, then splitting will
again occur due to the component of angular momentum M x , which
does not exist simultaneously with M z . The number of splitting com
ponents is again equal to 2 I + 1, since it is determined by the maximum
angularmomentum projection I. This quantity cannot depend upon
the direction of the magnetic field, and is related only to the atomic
states in the original beam.
Sec. 30] EXPANSIONS INTO WAVE FUNCTIONS 307
The eigenfunctions of M x are
", (30.17)
where I ^ fcj ^ I, and co is the angle of rotation about the #axis.
Functions (30.16) and (30.17) do not coincide, which is a natural
consequence of their being functions of noncommutative operators.
As a result of magnetic splitting in a field directed along the aaxis,
a beam with given value of k is split into 2 I + 1 beams with definite
values &j. Hence, the function (30.16) will be represented as the super
imposition of functions (30.17):
/
* (*) = 2>i +<*i> < 30  18 )
I*!
The square of the modulus  c kl  2 is proportional to the intensity
of the beam of the given projection M x = hk l obtained as a result of
the secondary splitting of the beam with a given M z .
Averages in quantum mechanics. Let us now find the average of X
in a state given by the wave function fy (x), represented in the form
of a sum (30.8). By definition the mean value is
XJTx^, (30.19)
x
i.e., the sum of possible values of X multiplied by the corresponding
probabilities. Let us substitute here u\ from (30.13) and cj from (30.11) :
)ty*(\x)dx. (30.20)
We shall now replace the product X^* (X, x) by X* ^* (X, x) and we
first sum and then integrate. Then we obtain
X = (*)JTcXX* 4* (X, x) dx . (30.21)
x
But the operator X* does not depend upon any definite value of X
(for example, if X = p x , then X* =  sg ).
Therefore, X* stands outside the sign of the summation :
X = J\J, (s)X*27<4** (X, x) dx . (30.22)
x
The sum c* fy* (X, x) = <i* (x) , since this is an equation which is a
x
conjugate complex of (30.8). Therefore,
20*
308 QUANTUM MECHANICS [Part III
X = <l> (x) X* ty* (x) dx (30.23)
or, from the Hermitian condition for the operator X (30.2), (30.3),
X = ^ * (x) X <]> (x) dx . (30.24)
Thus, in order to calculate the mean value of X in a state fy (x),
it is not necessary to know the eigenvalues of X, since it is sufficient
to calculate the integral (30.24).
The eigenvalues of the square of the angular momentum. If fy (x)
is one of the eigenfunctions of the operator X, then the mean value X
is simply reduced to this eigenvalue. Indeed, then
(X, x)\*dx = X.
Taking advantage of the foregoing remark, it is easy to calculate
the mean value of the square of the angular momentum.
First of all it may be noted that in the SternGerlach experiment
the mean values of the squares of all three angularmomentum pro
jections must be the same, because it is absolutely immaterial what
the notation of the coordinate axis is along which the magnetic field
is directed:
W x = W y ^~Ml. (30.25)
It follows that the mean value M 2 is equal to three times the mean
value Ml\
In the original beam all values of M z ~hk from hi to hi are equally
probable. This means that Jf is equal to
whence
l&* = h z l(l+ 1). (30.28)
But it was shown in Sec. 29 that M 2 is commutative with M Zl so
that M 2 and M x exist in one and the same state. In the SternGerlach
experiment the atoms in the beam occur predominantly in the ground
state. This state is characterized by a certain absolute value of the
angular momentum. Therefore, the mean value of the angular momen
tum in such a state is equal to its eigenvalue
Jf2 = W^ Wl (I + 1) . (30.29)
Sec. 30] EXPANSIONS INTO WAVE FUNCTIONS 309
The result (30.29) may appear somewhat surprising because the
eigenvalue of the square of the angular momentum is equal not to the
square of its greatest projection A 2 / 2 , but to some greater amount.
However, if M \ were equal to h 2 1 2 , i.e., its greatest value and M 2 = h 2 l 2 ,
then for the remaining projections there would remain an identical
zero. The other projections cannot have any definite values, including
zero values, at the same time as M z ^ 0. Therefore, the square of the
angular momentum is somewhat greater than the square of the maxi
mum value of any of its projections. The only exception is when all
three projections are equal to zero (see Sec. 29).
Composition of angular momenta. Knowing the absolute value of
the angular momentum, we can now indicate a rule for the composition
of the angular momenta of two mechanical systems. Let the greatest
angularmomentum projection of one system equal hl^ and that of
the other system AZ 2 ; and also let l^l^. Then the projection of the
smaller angular momentum in the direction of the larger one is con
tained between hl 2 and &Z 2 , which, when added to the larger angular
momentum, yields values ranging from A(Z 1 + / 2 ) to h (l Z 2 ). It
follows that the greatest projection of the resultant angular momentum
upon any arbitrary direction in space is equal, in units of /&, to
ZZi + ^2, h + li 1, h + k 2 >.> li~l z . (30.30)
The eigenvalues of the square of the sum of the angular momenta are
+ w (*i + h + 1) , A 2 (/ x + z,  1) & + y , ... ,
The rule for composition of angular momenta formulated here agrees
with the result that the value of a vector sum is contained between the
sum and the difference of the absolute values of the vectors.
Quantum equations of motion. Let us suppose that a certain opera
tor X is given. It is required to find the operator form of its total time
derivative, i.e., X. We shall first of all determine the total derivative
of the mean value X. In accordance with (30.24), for any state with wave
function d, this derivative is
Let us substitute here the derivatives  and ~ from the
ot ot
Schrodinger equation (24.11), whose righthand side we shall represent
as & ^, where & is the Hamiltonian operator [see (29.7)]. From this,
310 QUANTUM MECHANICS [Part III
We transform the first integral on the righthand side in accordance
with the Hermitian condition for $ ', namely
We now combine all three integrals and obtain
X = JV (} + J [& X  X Jf]) <; dx . (30.31)
If we now define the operator X by the equality
X ^ (fy*i<\>dx, (30.32)
then we obtain the equation
* /)C A ~ >S /V
X   + [X X  X Jf] . (30.33)
The operators of linear momentum, angular momentum, and coordi
nate that have been employed up till now do not depend upon time
explicitly. For them, only the second term of (30.33) remains:
X r "i/P "i 1 f y(P~\ t Q C\ O A \
= 7 LeJ^T A Ae^TJ . ^oU.OTcj
h
Thus, if a given operator commutes with the Hamiltonian operator
Jf 7 , then X = 0. It is then natural to call the quantity X a quantum inte
gral of motion. In accordance with the general result of Sec. 29,
quantum integrals of motion have a common state with energy, since
their operators are commutative.
We shall now find the equations of motion for the x and p x opera
tors. From (29.7), the energy operator $ is equal to ~ + U. Here,
&tfY\i
only PX is noncommutative with x ; for p x we find
Px X X p x = p x X p x X PX ~ p x X p x X p x
2h
= p x (p x X X p x ) + (p x X X px) p x = PX .
It follows that
i.e., the operator x is related to the operator p x by the same expression
as the quantities x = v x and p x in classical mechanics.
Let us now find p x . p x does not commute with U. The commutator
of U and p x is easily evaluated :
Sec. 30] EXPANSIONS INTO WAVE FUNCTIONS 311
whence, symbolically,
p x U^r (30.36)
Hence,
aft
p x =yL, (30.37)
which is completely analogous to the classical relationship between
the momentum derivative and the force.
The quantum equations of motion (30.34) were the starting point
for W. Heisenberg, who arrived at quantum mechanics independently
of Schroclinger. The equivalence of both approaches was shown some
what later.
The wave function and measurement of quantities. The probability
amplitudes characterize the properties of a system in relation to the
results of measuring certain quantities. If a system occurs in a state
with wave function fy (x), and the quantity X is measured, then the
probability of obtaining the given value of X is [see (30.11), (30.13)]
For example, in the SternGerlach experiment, the particles in the
original beam have all angularmomentum projections between hi
and hi. Measurement results yield 2 Z + l beams, each of them corre
sponding to the 25th angularmomentum projection given by a definite
value hk. However, the same measurement in a field directed along the
#axis of the original system would split the beam according to the
xth angularmomentum projections. Both angularmomentum pro
jections do not exist simultaneously, and the initial states of the
particles in the beam were identical. It follows that, as a result of
measurement, the particles occur either with a definite zth, or with a
definite xth angularmomentum projection.
A measurement of a microscopic entity essentially changes the state
of the latter. This is the fundamental difference between the concepts
of measurement in classical and in quantum physics : a classical meas
urement has an infinitesimaJly small effect on the object being
measured.
As a result of measurement, the angularmomentum projections in
the original beam acquire 2 1 + 1 values, no matter how the measure
ment is performed. The state of these particles after measurement is
essentially different and depends upon how the measurement was per
formed. But by performing measurements of a large number of iden
tical entities, we can find out in what state they were before measure
ment, quite independently of the method of measurement. For this
reason, a quantum measurement yields physical results which are
just as objective as those given by a classical measurement though,
obviously, within the limits permitted by the uncertainty principle.
312 QUANTUM MECHANICS [Fart III
Thus, in the SternGerlach experiment it appears that the particles
had an absolute angularmomentum value M 2 = h 2 l (Z + l), while
the direction of the angular momentum in space was arbitrary (a non
polarized beam).
The repeated measurement of the 2th angularmomentum projec
tion, in the beams which had passed earlier through a field directed
along the zaxis, gives a definite value of M 2 =h 2 l (I + 1) and a definite
value of M z hk.
Exercises
1) Expand the function <> = , in an infinitely deep rectangular
\/a
potential well, in terms of functions (25.12).
a a
f , , ^ V^ f . n(n + l)x
c n = dub,, d x I sin dx =
J a J a
2) Find the energy eigenvalues for a symmetrical quantum top. The energy
of the symmetrical top is
Introducing M 2 , we have
1
Substituting the eigenvalues for the angular momentum and its projections,
we at last find
Sec. 31. Motion in a Central Field
The motion of an electron in a central attractive field is the princi
pal problem in the quantum mechanics of the atom. And it is not
necessary to regard the field as strictly Coulomb in character. For
example, in alkalimetal atoms, an outer electron which is bound
relatively weakly to the nucleus moves in the field of the nucleus and
the socalled atomic residue (i. e., all the other electrons). The charge
density distribution for these electrons possesses spherical symmetry
and therefore produces a central field. We shall suppose that the poten
tial energy of the electron is equal to U (r), where r is the distance
from the nucleus.
The energy operator and the angularmomentum integral. The equa
tion for the energy eigenvalues of an atom (24.22) is, as usual,
__ _ A<J, _j_ [7<j, g^ m (31.1)
Sec. 31] MOTION IN A CENTRAL FIELD 313
Here, m is the reduced mass of the nucleus and the electron, which
mass is very close to the mass of the electron. Since the field is central
we must pass to spherical coordinates. The Laplacian operator in
spherical coordinates was obtained in Sec. 11 (11.46). Using this
expression, we rewrite (31.1) explicitly:
_jfrLLLA r ii*. . _L/L d sina** i * d ^
2m Lr 2 dr T dr ^ r* [sin* db * ^ d* ^ sin a # 2 9 2
<l>. (31.2)
The operator involving angular differentiation is simply the square
of the angular momentum introduced by us in exercise 4, Sec. 29.
Therefore, equation (31.2) can also be rewritten as
It follows that the Hamiltonian operator <& [see (29.6)] is related to
the angularmomentum operator in the following way:
_
2m
I/ (r) . (31.4)
v ' v 7
Reducing to an ordinary differential equation. The operator M*
involves only the angles & and 9 and derivatives with respect to them.
All the derivatives with respect to angles in the operator $f are con
tained in the one term If 2 , while all the remaining terms involve only
r and the derivative with respect to r. Consequently, the operators 3t?
and M* are commutative, since M 2 commutes with any function
of r and, of course, with r itself. Commutative operators have eigen
values in the same state. Therefore, in a central field, the square of the
angular momentum and one of its projections have (together with
energy) eigenvalues, which, in accordance with (30.34), are quantum
integrals of motion. All the other quantities which are not integrals
of motion do not exist in the same energy state (in classical mechanics
they, naturally, exist but are not conserved).
Thus, in equations (31.3) and (31.4), we can substitute in place of
M 2 its eigenvalue h 2 l (I + 1) from (30.29). Then any angular dependence
will be eliminated from equation (31.3) and, in place of the partial
derivative with respect to r, we will get the total derivative :
It is considerably more simple to solve this equation than the partial
differential equation (31.2). The form of (31.5) corresponds to (5.6)
in classical mechanics, where it was also possible to eliminate all
^variables except r with the aid of the angularmomentum integral.
314 QUANTUM MECHANICS [Part III
Reduction to onedimensional form. It is convenient to reduce
equation (31.5) to a onedimensional form. To do this the treatment
is similar to that used in the problem of the propagation of spherical
waves [cf. (19.6)] we introduce the function
y = ri } tli . (31.6)
t T y
Without repeating the computations by means of which the one
dimensional form (19.6) was obtained, we write down the analogous
equation for /:
The wave function at large and small distances from the nucleus. As
long as the form of U (r) has not yet been made definite, we can con
sider (31.7) only in two limiting cases: for very large and for very
small distances from the nucleus.
The field of the atomic residue is not effective at very small dis
tances from the nucleus, and there remains only the Coulomb rela
tionship 1 [7 =  (Z is the atomic number of the element).
However, if r is very small then the term ~ ^ ty is, in any case,
larger than the term Ufy, which involves r in the denominator only
in the first degree, and all the more greater than &k. Hence, in direct
proximity to the nucleus, the wave equation is of very simple form :
In this form it is solved by the substitution
X r, (31.9)
so that
a (a I) /(/ + !). (31.10)
This equation has two roots:
a = / + 1 and a =  Z . (31.11)
But the second root gives fy = r~ l ~ l from (31.7); at the point r = 0,
this function of ^ becomes infinite for all L Therefore, we must discard
the root a = I and take the relationship between ^ and r for small r
in the form
^==^ = H*. (31.12)
* The result (31.12) is true for I = as well, even though the term + a
in this case does not exist at all and cannot exceed U (r) fy.
SeC. 31] MOTION IN A CENTRAL FIELD 315
The greater the angular momentum, the higher the order of the
wavefunction zero at the coordinate origin. Only for Z = does it
remain finite close to the nucleus. This can be understood by analogy
with classical mechanics: angular momentum is the product of mo
mentum by the "arm," i.e., by the distance from the origin; Z =
corresponds to a zero "arm" and a zero angular momentum. There
fore, there is a nonzero probability of finding the electron at the
origin. In the old version of quantum mechanics (due to Bohr), the
electron orbit with zero angular momentum passed through the
nucleus. The larger angularmomentum values correspond to larger
"arms" and, correspondingly, in quantum mechanics, to a smaller
probability of finding an electron close to the nucleus.
The behaviour of the wave function close to the origin can also be
explained as follows. A centrifugal repulsive force acts on the particle ;
to this force there corresponds an effective potential energy ^ ~ .
This energy limits the classically possible region of motion for small r.
In quantum mechanics the particle penetrates the centrifugal barrier,
though more weakly the greater r, i.e., the higher the barrier. There
is no barrier for I = and there is nothing to prevent finding the par
ticle at the origin.
The terms ~ ^~ fy and Ufy must be discarded for large r in the
wave equation, because U (r) is assumed to be zero at infinity,
U( 00)^0. Then the equation is also greatly simplified:
Its general solution appears thus:
x = C ie * +C 2 e * . (31.14)
Positive and negative energy values. We consider two cases. Let the
energy be positive, <?>0. Here, x appears as follows:
X = (7 1 c h + C 2 e h . (31.15)
Both terms remain finite for any value of r. Therefore, two constants,
C : and <7 2 , must be retained in the solution. We came across the same
situation in considering the solution of wave equation (25.33) for a
potential well of finite depth.
Any general solution of a secondorder differential equation involves
two arbitrary constants. Let us suppose that the solution (31.12),
which holds for small r only, is continued into the region of large r,
where it is not of the simple form r 1 , but nevertheless satisfies the
316 QUANTUM MECHANICS [Part III
precise equation (31.7). A certain integral curve is obtained for this
equation. But any integral curve can be represented by properly
choosing the constants in the general solution. As r tends to infinity
this solution acquires its asymptotic form (31.15) if <f>0. The ex
pression (31.15) remains finite when r>oo for any constants C l and
C 2 . It follows that, for a positive energy, the wave equation always
has a finite solution for any values of r. Therefore, the values for $ >
correspond to a continuous energy spectrum, since the wave function
satisfies the required conditions at zero and at infinity for any S > 0.
In accordance with (31.15), the probability of finding an electron at
infinity for r>oo does not become zero; i.e., this case corresponds
to infinite motion, as in the classical problem considered in Sec. 5
(see also Sec. 25).
Thus, the general rule has been confirmed that infinite motion
possesses a continuous energy spectrum.
Now let <?<0 or S = \ S \ . Then (31.14) must be represented
thus:
_ _
x = C r 1 e"" h " + C 2 e " ". (31.16)
Here the first solution tends to infinity together with r and we must
therefore put C : = 0, so that x will involve one instead of two arbitrary
constants :
* = C 2 e * . (31.17)
The condition for eigenvalues. If we now draw an integral curve
from the coordinate origin, proceeding from (31.12), then, as a rule,
for large r it will not be reduced to the form (31.17). For all negative
energy values, except certain ones, the integral curve is represented in
the form (31.16) at infinity when <f <0 and, hence, does not satisfy
the boundary condition imposed on the wave function. Only for those
energy values for which it turns out that
C 1 (<?)=0 (31.18)
does the wave equation have a solution. This corresponds to a discrete
energy spectrum. At the same time, x (<*>) becomes zero, so that the
finite motion has a discrete energy spectrum, as expected.
The Coulomb field. The transition to atomic units. We shall now
find this spectrum for an electron in a purely Coulomb field:
U(r) = ?. (31.19)
This occurs in a hydrogen atom (though not in a molecule!), in
singly ionized helium, doubly ionized lithium, etc. Z, as usual, denotes
the atomic number of the nucleus.
Sec. 31] MOTION IN A CENTRAL FIELD 317
The wave equation (31.7) is now written as
__
2m dr*
We have straightway taken the case of negative energies that leads
to a discrete spectrum.
It is convenient here to change the units of length and energy
similar to the way it was done in the problem of the harmonic oscillator
(Sec. 26). In place of the CGS system (where the basic units are the
arbitrary quantities centimetre, gram, second) we take the following
units : the elementary charge e, the mass of the electron m, and the
quantum of action h. From these quantities we form the unit of length
h * =5.29 17 x 10 9 cm
me 2
and energy
Hence, if we put e = l, m = l, h = l in equation (31.20), then length
and energy will be measured in these units. Let us call this length :
and energy s:
* /Ql 90\
(31.22)
so that, of the constants, the wave equation will involve only the
atomic number Z:
(31.23)
Solution by the seriesexpansion method. We look for the solution
of this equation in the form of a series expansion. We shall proceed
here from the solutions obtained for large and small values of i;
(i.e., r).
In accordance with equations (31.12) and (31.17), we write / in
following form:
< SL24 >
n =0
The first factor determined the form of x for ^>0, the second factor
should basically correspond to the form of x for large , and the series
interpolates, as it were, between the limiting values.
318 QUANTUM MECHANICS [Part III
Differentiating (31.24) twice, we obtain
00
(31.25)
The first term on the right is simply 2 e^. Hence, it cancels with the
same term in (31.23). We group the remaining terms so that in one of
them the degree of is everywhere less by unity than in (31.24) and,
in the other, less by two units. In addition we eliminate the common
factor e5^ e . We shall now have an equality between two such
series :
00
[1(1+ 1)  (n + I + 1) (n + I)] X^H'^
(31.26)
An equality between series is possible only when the coefficients of
the same powers of ^ coincide. On the lefthand side the power M + /
will have a coefficient involving
Hence
 (n + / + iT^TT+ 2)~ ' .
Examining the scries and the condition for eigenvalues. From the
relationship (31.27), all the coefficients x are determined consecu
tively. We must neglect the constant numbers I and Z in equation
(31.27) when n are large; there then remains the limit
We met with a similar expression in the problem of the harmonic
oscillator (26.16). In the case of large it reduces the whole series to
an exponential form:
(31.29)
But such a series cannot give a correct solution to the wave equation
because, if we substitute (31.29) in (31.24), we obtain ^(oo) = oo
despite the boundary condition. However, if all the coefficients become
zero from a certain x+i onwards, the series (31.29) degenerates to a
polynomial. Then, being multiplied by eS^^r, ^ gives ^ (oo) = 0,
Sec. 31] MOTION IN A CENTRAL FIELD 319
as expected. It can be seen from (31.27) that &, + 1 * s equal to
zero if
Z (n + 1 + 1) V2? = 0, (31.30)
Finally, going over to conventional units and taking into account
the sign of the energy, we obtain the required spectrum:
T (31.32)
Quantum numbers. The number n is the degree of the polynomial
[ n (it is called the SoninLaguerre polynomial). A more detailed
analysis shows that this polynomial becomes zero exactly n times,
corresponding to its degree. Therefore, if we examine the dependence
of the wave function on radius, it has n zeros or "nodes/* not counting
the zero at r = oo and at r = Q, which all functions with Z^O have.
The term node instead of zero is given by analogy with the nodes of a
vibrating string fixed at both ends. In future we shah 1 call n r the degree
of the polynomial and denote by the letter n the whole sum
n==tt r + Z+l. (31.33)
It is convenient to use these quantities also in the more complex
cases of many electron atoms. Even though the energy in such a case
does not have the simple form (31.32), the numbers n, n r , and I are
convenient for classification of the states.
I is called the azimuthal quantum number. As we know, it defines
the angular momentum of an electron. The following system of nota
tion is used in spectroscopy : the electron state with I = is called the
sstate and, corresponding to I = 1, 2, 3, we have the p, d and /states.
There are no greater values of I in nonexcited atoms. Combining the
angular momenta of separate electrons according to the rule of vector
addition (30.30), we obtain the angular momentum L of the atom as a
whole. The states with L = Q, 1, 2, 3 are termed S 9 P, D, F, while
states with greater L are named by subsequent letters of the Latin
alphabet.
fc, [see Eq. (29.21)], i.e., the angularmomentum projection on some
axis in units of A, is called the magnetic quantum number, since the
external magnetic field is usually directed along this axis.
n r is the number of wavefunction zeros as related to the radius
(for r^O and r=^oo) and is called the radial quantum number.
Finally, the sum (31.33) is called the principal quantum number.
In accordance with (31.32), the binding energy of an electron in a
hydrogen atom is
320 QUANTUM MECHANICS [Part III
An analogous expression is obtained also for the positive helium
ion. Apart from the difference of Z 2 = 4 times, there is a more subtle
difference due to the fact that the reduced mass of the helium atom
differs somewhat from the reduced mass of a hydrogen atom as a result
of a difference in the nuclear masses.
The state with n=l is the ground state. The atom cannot emit
light in this state because it is impossible to make a transition to a
lower state. For more detail about radiation, see Sec. 34.
The parity of a state. The state of an electron in an atom is character
ized by one more property, which (as opposed to energy and angular
momentum) does not correspond to any classical analogue. This is
the parity of a wave function with respect to coordinates.
To begin with let us consider the wave function of a separate electron.
The wave equation (31.1) does not change its form if we substitute
x =  of y,   */', z =  z'. (31.35)
This transformation is termed inversion : it transforms a righthanded
coordinate system to a lefthanded one. No rotation in space can make
these systems coincide (like lefthand and righthand gloves) (see
Sec. 16).
The wave equation (31.1) is linear. Therefore, if it has not changed
its form, then its solution (determined by the boundary conditions
within the accuracy of the constant factor) can acquire only a certain
additional factor:
', ',*') (31.36)
But, in principle, the primed lefthanded system differs in no way
from the unprimed, righthanded system. For this reason, the trans
formation of inversion must involve the same transformation factor C:
x,v,z). (31.37)
Substituting this in (31.36), we obtain
whence
C 2 = l, G=l. (31.38)
The function is termed even for (7 = 1 and odd for C= 1. The
eigenfunctions of a linear harmonic oscillator possessed an analogous
property; here the* energy operator was also even, jf? (x) = jf?( x),
while the wave functions alternated depending upon the eigenvalue
number n (i.e., they were either even or odd).
Parity and orbital angular momentum. Let us now find out what
it is that determines the parity of a wave function in a central field.
Sec. 31] MOTION IN A CENTRAL FIELD 321
To do this, it is convenient to utilize its form near the coordinate
origin:
fy = r l . (31.39)
In order to find the angular dependence of the wave function as
well, it is sufficient to investigate it to the approximation that yields
equation (31.39) since the terms U and S , which do not depend upon
the angles, are thereby discarded. The angular dependence of the
solutions of the precise and shortened equation is the same. This
shortened equation is, obviously, simply the Laplace equation
A^ = 0. (31.40)
Equation (31.40) is satisfied by a homogeneous polynomial in x, y, z
<p = x l + ax 1  1 y + ... + bx l  k ~ m y k z m + . . . (31.41)
of degree I for certain relationships between its coefficients a, . . . ,
b, ... .It is clear that the degree I of this polynomial is equal to the
degree I in equation (31.39). But the degree I of (31.41) defines the even
or odd nature of ^ with respect to the inversion (31.35). It follows that
wave functions with even orbital angular momenta I are even, and
those with odd orbital angular momenta I are odd. In a multielectron
atom, the total parity of the wave function is determined by the parity
of all the wave functions for the separate electrons (this by no means
signifies that the wave function of an atom is equal to the product of
the wave functions of the separate electrons!). Therefore, the parity
of the total wave function is equal to the parity of the number JE7fc
i
where /, are the orbital quantum numbers of the electrons. As we know,
the total angular momentum of the atom is equal to the vector sum
of the angular momenta of its electrons.
Parity as an integral of motion. We shall explain the significance of
the parity of a wave function. To begin with, let us point out that the
inversion (31.35) can be represented by an operator G such that
Gty(x 9 y,z) = ty(x,y,z). (31.42)
Since the Hamiltonian operator in an atom & is an even function of
coordinates, we can write
&<$ = &. (31.43)
Whence it follows that the parity operator is commutative with the
Hamiltonian operator
G3fty = jeG<l>. (31.44)
The eigenvalues of the operator G are the numbers G = I (31.38),
because
Gfy = ty ( x,  y,  z) = C<Jj . (31.45)
21 0060
322 QUANTUM MECHANICS [Part III
According to (.31.44) and (29.28) these numbers exist simultaneously
with the energy eigenvalues.
We shall now consider what limitations can be imposed, by the
law of conservation of parity, on possible transitions in the atom.
Suppose we have an excited multielectron atom with total angular
momentum L = 0, i. e., in the $state. Then let there be in this atom
selectrons and an odd number of pelectrons. Consequently, the atom
is in an odd state. Let the excitation energy be sufficient for the atom
to emit one of the ^electrons, so that after the rearrangement of the
electron cloud the atom remains in the $state with = 0. Since angu
lar momenta are combined vectorially, such a state may result both
for an odd and an even number of ^electrons. According to the law
of conservation of total angular momentum, an electron may be emitted
only with an angular momentum equal to zero because, according to
assumption, the angular momentum of the rest of the system is equal
to zero before and after the transition. It follows that the electron can
be emitted only in an even sstate.
After the emission of the electron, an even number of ^electrons
remains in the ion, and the emitted electron is also found to be in an
even state. But this is impossible since the initial state was odd and
the final state was even, the total energy being constant. Hence, the
laws of conservation of parity and angular momentum may exclude
transitions which are permissible energetically. We have considered
a typical case of a transition which is "forbidden" by parity selection
rules (Z/ into Z/ with changed parity).
The law of conservation of parity by no means follows from the
law of conservation of angular momentum, since parity depends upon
the arithmetic sum of I while total angular momentum depends upon
the vector sum.
The law of conservation of angular momentum in quantum mechan
ics must always be used together with the law of conservation of
parity. In origin, these laws have a common basis: they both follow
from the invariance of equations with respect to the orientation of
coordinate axes in space. But all possible orientations are not exhausted
by axis rotations alone : an additional transformation is inversion which
is not reduced to any rotation. It is this that yields the parity conser
vation law in addition to the law of conservation of angular momentum.
In this form the parity conservation law can be unconditionally
applied to those systems in which electromagnetic interactions occur.
The considerably weaker interactions which occur in certain ele
mentaryparticle conversions probably satisfy a modified parity con
servation law (see Sec. 38).
Hydrogenlike atoms. Alkalimetal atoms somewhat resemble the
hydrogen atom. The outer electron in these atoms is relatively weakly
bound to the atomic residue, which consists of the nucleus and all the
remaining electrons. The wave functions for electrons of the atomic
Sec. 32] ELECTRON SPIN 323
residue differ from zero at smaller distances from the nucleus than the
wave function for the outer electron, so that the residue screens, as it
were, the nuclear charge. The field in which the outer electron moves
is approximately Coulomb, provided only that it is not situated in
the region of the residue. It is for this reason that the spectra of alkali
metal atoms resemble the hydrogenatom spectrum. The energy levels
of these atoms, which are due to excitation of the outer electron, are
given by the equation
& me* 1 _ /01 Aa .
*"'= W [n + A (I)]*"' < 3L46 )
where the correction A (I) depends upon the azimuthal quantum
number. It accounts for the deviation of the field from a purely Cou
lomb one at small distances from the nucleus.
Thus, the energy levels in alkali metals like the energy levels of
all atoms depend upon n and I. An exception is the hydrogen atom,
where the energy depends only upon n\ this is a special property of a
purely Coulomb field. For example, when n = 2, the azimuthal quan
tum number can take on two values: 1^0 and 1 = 1, while the corre
sponding energy levels of the hydrogen atom are close to each other
(the splitting of these levels is due to relativistic corrections to the wave
equation).
Exorcise
Construct and normalize the wave functions in a hydrogen atom with
00
1 = 0, 1, 2 and w= 1, 2, 3. Take advantage of the fact that j e~*x n dx = n !
Sec. 32. Electron Spin
The insufficiency of three quantum numbers for the electron in an
atom. From equation (31.34) the ground state of a hydrogen atom has
a principal quantum number n equal to unity. For n = 1 the azimuthal
quantum number I and the radial quantum number n r must be equal
to zero, since n = n r + l + 1, and n r and I can in no way be less than
zero. The ground state of a hydrogen atom is the sstate. The orbital
motion of an ^electron does not produce a magnetic moment because
the magnetic moment is proportional to the mechanical moment.
Yet, if the SternGerlaeh experiment is performed for atomic hydro
gen, the atomic beam will split, but only into two parts. However,
when Z = 0, as we have already said, there should be no splitting due
to orbital angular momentum, while for Z 1, the beam should split
into 2 I + 1 = 3 beams corresponding to the number of projections k
of the angular momentum I ( 1, 0, 1).
The same results if, instead of hydrogen, we take an alkali metal.
The electron cloud of any alkali metal consists of an atomic residue in
21*
324 QUANTUM MECHANICS [Part III
an /Sstate, i.e., one lacking an orbital angular momentum and one
electron in the sstate. In this sense, aklalimetal atoms resemble the
hydrogen atom. For this reason, the state of the atom is not described
by the three quantum numbers n, I, and k.
Intrinsic angular momentum or electron spin. Splitting into two
beams can be accounted for only by an angular momentum whose
greatest projection is equal to hj 2. Then it has only two projections
h/2 and h/2.
The SternGerlach experiment was given only as an example. In
fact, not only this experiment, but the whole enormous aggregate of
knowledge about the atom indicates that the electron possesses a
mechanical moment h/2 that is not related to its orbital motion. This
mechanical moment is termed the spin. It can be said that an electron
is somewhat reminiscent of a planet which has an angular momentum
due not only to its revolution about the sun, but also to rotation on its
own axis.
The analogy with a planet is not farreaching since the angular mo
mentum of rotating rigid body can be made equal to any value, while
the spin of an electron always has projections A/2 and no others.
Therefore, spin is a purely quantum property of the electron; in the
limiting transition to classical mechanics it becomes zero. We must not
take the word "spin" too literally, for the electron actually does not
resemble a rigid body like a top or a spindle.
Spin degree of freedom. The analogy between an electron and a top
consists in the fact that their motion is not described by their position
in space alone, but possesses an internal rotational degree of freedom.
There is a certain analogy between the electron and the light
quantum. As was shown in Sec. 28, in addition to its wave vector,
the state of a quantum is described by a polarization variable which
takes on two values. Similarly, the electron has, in addition to its
spatial coordinates, a spin variable a which assumes two values (since
spin has only two projections).
Spin operators. When we write ^ (#) we have in mind the whole group
of values of the wave function for all x, i.e., fy at all points of space.
The action of an operator on fy (#) denotes a linear transformation
of ip in the whole space, since, in accordance with the superposition
principle, all operators in quantum mechanics are linear. Taking into
account the spin variable, one has to write fy (x, a), where or takes on
only two values. The action of the spin operator on fy (x, a) denotes
the replacement of fy (x, 1) by some linear combination of ip (x 9 1)
and fy (x, 2) ; the action of the operator on fy (x, 2) is determined anal
oguously. Linear operators depending upon a can denote nothing
other than a linear substitution as applied to a function of "two
points" a = l and a 2.
We shall try to determine, in explicit form, how spin angular
momentum projection operators should act upon functions of the
Sec. 32] ELECTRON SPIN 325
spin variable a. The following requirements are to be imposed
on them.
1) The eigenvalues of all three spin projections must be equal to
fc/2.
2) The same commutation rules (29.33) (29.35) must exist for
them as for the components of orbital angular momentum, otherwise
the sum of the orbital and spin angularmomentum operators will not
possess the property of angular momentum.
3) For the same reason we must require that the spinprojection
operators should be Hermitian.
4) In coordinatesystem rotations, spinprojection operators must
behave in the same way as vector components so that the commuta
tion rules for these operators, in a rotated system, should not differ
from the rules of the original system in which the operators were de
fined.
Corresponding to these requirements, W. Pauli found the required
operators, which we shall now form.
We shall write the group of functions fy (x y a) in columns instead of
rows; the meaning of fy (x, ci) does not thereby change, of course.
In addition, for brevity, we shall omit the coordinate dependence con
tained in the argument x. Thus, fy (<*) denotes the column
p *
Here, each component satisfies Schrodinger's coordinate equation
(24.22). In the most general case the action of a linear operator on the
function (32.1) reduces it to the form
As we know, one of the angularmomentum projections can always
exist together with the square of the total angular momentum, since
the substitution rules for spin components are the same as for orbital
angular momentum (condition 2). To be specific, we shall consider
that there exists the 2th projection a*. If the operator a z has an eigen
value in the given state, its application leads to the multiplication of
the function (32.1) by some number without mixing of the components.
This number is equal to A/2, depending upon what the sign of
the spin projection a z is in the given state. Let the function <p (1)
be multiplied by +h/2 and the function <p (2) by A/2, so that
326 QUANTUM MECHANICS [Part III
The equality sign between columns denotes a line by line equality
of the expressions, i.e., S*<p (1) = y ^ (1), c z (2) = ^fy (2) . The
form of the functions corresponding to various spin projections can
immediately be seen : to the projection h/2 there corresponds a function
( ) ' w ^ e ^ ne function I , , 2 \) corresponds to the projection A/2.
The first of them, if we substitute it in (32.2), is entirely multiplied
by A/2, while the second is multiplied by A/2, since the change of
sign for the zero component of the function does not signify anything.
The operators a x and <r y cannot have eigenvalues in these states.
It follows that they must in any case also interchange the components
of the wave function that we have defined, and not merely multiply them
by numbers. Simple multiplication operators would be commutative
with G Z . Once the form G Z is given, we can also determine the operators
for the other two components.
Let us temporarily go over to atomic units (see Sec. 31), i.e., we
put A ~ 1. We shall look for the operator 5 X (acting on the two functions)
in the most general possible form*:
(32.3)
In other words, we suppose that it replaces <pi by a^ + p^g and
^2 by Y^i + ^2* We act on c x fy with a*. Then, by definition of G Z
(32.2), we must change the sign in the lower row and divide both
rows by two :
If we act upon G z fy with G X , then we must first of all put a minus sign
in front of <ji 2 and divide both components by two, and then substitute
them in (32.3). This will yield
From (29.35), the difference a z <*x <y*S* must be equal to ia y in
atomic units :
__ y . (32.4)
Thus, the operator <r y interchanges the functions <pi and ^ 2 and multi
plies them by p and y, where p and y appeared in the definition
(32.3) for 5x.
* The argument o = 1 and a = 2 will in future be replaced by the index.
Sec. 32] ELECTRON SPIN 327
But if we proceed from or y , defining it analogously to (32.3), it will
turn out that G X also interchanges functions, i.e., it does not contain
the coefficients a and 8. Therefore we obtain
We now form the difference G X <r y <r y G X . First acting with G X upon
Gyfy and then with <y y upon a x fy, we equate their difference to i a z fy [see
(29.33)]:
whence it follows that
Y= (325)
Hence, conditions (29.33)(29.35) lead to the following form for
G x and G y :
r \ t
(32.6)
The operators *, ffy and <r* must be Hermitian. We once again
deduce the Hermitian condition (30.3) for the operator G X , insofar as
the result of Sec. 30 related to a continuous variable #, while here
we consider a discrete variable G. Let us write similarly to (30. la)
and (30. Ib), with spin dependence in explicit form:
Summation with respect to a now corresponds to integration with re
spect to x. We multiply the first two equations by <p? andj <];, respec
tively, and the second two by ^ and 4> 2 > sum w ^h. respect to cr, and
equate the results utilizing the fact that G X is a real number :
As was shown in Sec. 30, this condition must be identically satis
fied for any two functions #* and fy. From this we obtain
3d* ?* + Xt* 17 *i = *i P* X a * + 4*2 4^1 Xi* (32.7)
But this equation can hold only if
328 QUANTUM MECHANICS [Part III
i.e., B = He' v . There remains an arbitrary phase factor e' v which
2t
we choose equal to unity. Thus,
< 32  8 >
< 32  9 >
We jiote three operator relations
G x Gy " Gy Gx = 5 ff* J CF* GTx = <7x 0^ = ~^~ CTy J
. (32.10)
GyGz~ OF* CTy = g CT* ,
which are directly verified by substitution, and also expressions for
the squares 3, SJ, and SJ, which are obtained when they act twice
upon 4"
1 AM 1 i
= = T*'
(32.11)
s^ = ^; a;**.
Hence, in accordance with condition (1), the eigenvalues of SJ,
SJ, and aj are j . Since each operator is commutative with its square,
we see that the eigenvalues of a*, &y, and G Z should equal the square
roots of the eigenvalues of their squares, i.e., 1/2. Naturally,
these eigenvalues of a*, 5 y , and a z only exist separately and by no
means simultaneously.
The vector properties of spin operators. In order to prove finally
that the operators a x , a y , and G Z possess the properties of angular
momentum components, we must be convinced that in coordinate
rotations they transform like vector projections, i.e., we must verify
that condition (4) is satisfied.
Let us suppose that a rotation occurs around the zaxis through
an angle co. Then we must prove that the operators
5i = Sr*cosco+S y sin<o,
A, A . , A lO^.l^j
a y a* sin co + ar y cos co ,
formed by analogy with the vector projections on rotated coordinate
axes possess the same properties (1) and (2) as the original operators
S x and 5 y . First of all we have
>) (G X cos co 
= GX cos 2 co + 5 y sin 2 co + (G X 5 y + a y G X ) sin co cos co .
Sec. 32] ELECTRON SPIN 329
It can be seen from (32.10) that
a x ay + SyS x =0 (32.13)
(and analogously for any pair of components). Further, a and aj,
operating on ^functions, act like numbers, i.e., they simply multiply
it by 1/4. But then this also means that S ' J = j (cos 2 co + sin 2 to) = 7 .
Thus, the first property of 3* and S' y is retained under the rotation
of the components.
We now form the difference a x S y G y S x :
GX GySyG x ~ ~ <*x COS CO SU1 CO + Sy SHI CO COS CO + SjcSy COS 2 CO
o 4 5coscosinco ajsincocosco + 5xSysin 2 co
O S X Sy SySx = lS Z . (32.14)
But Sz=$z since the rotation occurs about the zaxis.
Any rotation in space may be obtained by successive rotations
about three axes. Therefore, it was sufficient to show that the basic
properties of the operators are preserved under rotations about any
one of the three axes.
The total angularmomentum operator. If we now form the sum
of the operators
J X = M X + S X ; jy = M y + $ y ; f z = M z + 3 z , (32.15)
then it will possess all the properties of an angularmomentum operator.
Naturally, we could not have added the components of o to the
components of M if they both did not transform identically under
rotations of the coordinate system, since, otherwise, equations (32.15)
would be noninvariant with respect to the choice of the system of axes.
The vector j is called the total angular momentum of an electron.
If the orbital angular momentum of the electron has a greatest
projection I, then the greatest projection of j can equal 1 + ^ or
I 5 In the first case, we say that the spin and the orbital angular
momentum are parallel ; in the second case, we say that they are
antiparallel.
Spin magnetic moment. The spin of an electron, similar to its orbital
angular momentum, is associated with a definite magnetic moment.
But experiment shows that the ratio of spin magnetic moment to
mechanical moment is twice as great as for orbital angular momentum.
There is nothing paradoxical in this because the result (15.25) cannot
be applied to spin. At the same time we can deduce the spin magnetic
moment from the Dirac relativistic wave equation for an electron
(Sec. 38); in agreement with experiment, it is found to be
* = (32  16)
330 QUANTUM MECHANICS [Part III
Hence, the projection of jx a on any axis is
i. 32.17
The quantity JJL O is termed the Bohr magneton. This is a natural
unit of magnetic moment.
The ratio ^ q ^ =  is called the spin gyromagnetic ratio. It was
QZ 771 C
first discovered in determining the mechanical moment caused by
magnetization of iron rods (the Einsteinde Hass experiment). Spin
was not known at that time and it appeared strange that the gyro
magnetic ratio was not equal to ^  as follows from (15.25). It
is now known that magnetism in iron is connected with the spin
of certain of its electrons.
The fine structure of atomic levels. Spin magnetic moment interacts
with the magnetic moment of orbital motion and with the spin
angular momenta or other electrons, if the atom is of the multi
electron type. This interaction is proportional to the magnitude of
both magnetic moments, i.e., it involves the product of gyromagnetic
ratios. The latter is inversely proportional to c 2 and, hence, is an
essentially relativistic effect.
Electron velocities in atoms are everywhere small compared with
the velocity of light, with the exception of the internal regions of
the atoms of heavy elements. Therefore, a quantity involving c 2
in the denominator is usually small compared with other quantities
on the atomic scale; the interaction energy for magnetic moments
is less than the distance due to electrostatic interaction between
energy levels. As a result of the interaction between spin and orbital
angular momenta, the energy level of a separate electron corresponding
to a total electron angular momentum j = I + ^ differs a little (when
placed in a central field) from a level with a total angular momentum
j^l y*; this is because the angular momenta are parallel in
the first case and antiparallel in the second. But the energies of
two parallel and antiparallel angular momenta differ.
The only scalar quantity which is linear with respect to each of
two pseudovectors ji^ and fx 2 is \L (X 2 . Therefore, to the lowest approxi
mation, the interaction energy of two magnetic moments is pro
portional tO [JL 1 fJL 2 .
The spacing between the levels j = I + ^ and j = I is small
compared with that between electron levels with different I. Therefore,
a magnetic interaction contributes only a small splitting of the
* The angular momentum is determined by means of its greatest projection.
Sec. 32] ELECTRON SPIN 331
electron level with a given I into two levels. This splitting is called
the fine structure of the level.
We note that such a simple splitting into two levels takes place
for a separate electron in a central field, for example, for the outer
electron in an alkalimetal atom.
Isotopic (isobaric) spin. The splitting of an atomic level into two
levels with j = 1 + 1/2 and j = l 1/2 is due to weak magnetic inter
actions between spin and orbital magnetic moments. Since each
magnetic moment contains c in the denominator [see (32.16)], such
interaction is relativistic in nature and must vanish if the electrostatic
forces alone are taken into account. This means that the energies
of two states with spins parallel and antiparallel to the magnetic
moment coincide if magnetic forces are completely neglected.
An analogous situation exists in the domain of nuclear interaction.
The nuclear forces which hold nuclear particles (neutrons and protons)
together are not of electromagnetic origin. At least we have no in
dications that both types of force nuclear and electromagnetic can
be deduced in a unique manner from some first principle. As yet,
no experiment suggests that such derivation is at all possible. On
the contrary, there are many facts proving that nuclear interactions
are independent of the electrical properties of particles.
First, we have the socalled mirror nuclei. These are pairs of nuclei
which have all the neutrons interchanged with all the protons, and
vice versa. For example, H 3 consists of one proton and two neutrons,
and He 8 , of two protons and one neutron. All the main properties
of such nuclei are similar both qualitatively and quantitatively,
and the small differences that still do exist can readily be explained
by the difference in charge and magnetic moment of the neutron
and proton. We can therefore say that the substitution, in a nucleus,
of all protons by neutrons and all neutrons by protons leaves the
nuclear interactions invariant, i.e., the nuclear interaction between
two protons and two neutrons is the same if we neglect electro
magnetic forces.
Second, the scattering of neutrons and protons on protons indi
cates that the elementary nuclear interactions neutronproton and
protonproton are also the same. This is a stronger statement than
the previous one, because the interaction between two unlike particles
is also taken into account.
Comparing this situation with that in the atom, we can say that
there is no splitting of nuclear states if the strongest interactions
alone are considered; the actual splitting is due to the much weaker
electromagnetic interactions.
Let us, therefore, neglect for a time the weakest interactions.
We can then consider the neutron and the proton as two states of
a single particle the nucleon. These states do not differ in energy
like those of an electron with two different spin projections in the
332 QUANTUM MECHANICS [Part III
absence of a magnetic field. If such a field is switched off, both states of
the electron fall together in energy ; if all the electromagnetic interactions
are switched off, certain states of nucleon pairs fall together, too.
We have said that the spin can be considered as an internal degree
of freedom of the electron. It is reasonable to say that the electric
charge is the internal degree of freedom in the nucleon. Both degrees
of freedom assume only two values with a dichotomic variable cor
responding to them. It will be shown that there exists a farreaching
formal analogy between these degrees of freedom. Let us say that
the nucleon possesses (besides its usual, nucleon, spin) another "spin"
variable, which defines its "charge state." Like mechanical spin,
this variable assumes only two values. It is called the isotopic spin
(sometimes, and more consistently, the isobaric spin). We shall say
that the projection on some imaginable axis of isotopic spin is equal
to +1/2 which corresponds to a proton, the opposite projection
corresponding to a neutron. Some years ago, the reverse convention
was used, but this is immaterial. Now let us consider three nucleon
pairs : protonproton, neutronproton, and neutronneutron. According
to what we have already said, the first pair corresponds to a resultant
projection 1 of the isotopic spin, the second pair to a projection 0,
and the third, to 1. In the absence of electromagnetic forces, none
of the three projections split in energy.
But if these states coincide in energy they can be considered as
having a resultant spin 1 and differing in then: projections only.
Spin angular momentum 1 can assume just three projections; and
here we can say that the resultant isotopic spin angular momentum 1
has three different projections on some imaginable zaxis. In the
absence of electromagnetic forces, the physical choice of such a
"zaxis" is unimportant. Note, for comparison, that if no magnetic
field is applied to the electron, any direction in space is preferred
(for example, the zaxis).
Changes in projections of ordinary spin can be due simply to ro
tations of the coordinate frame. If no preferred directions in space
exist, such rotations are unlimited. Now we can consider different
isotopic spin projection as due to "rotation" of some frame also.
But this rotation is purely formal in nature and has nothing in common
with the rotation of geometrical space, except their mathematical
expressions. If the isotopic spin vector^? with components T*, T y , T*
is introduced, then its rotations are described exactly by the same
formulae as (32.12). The corresponding angle of rotation has no more
geometrical meaning than the axes which rotate.
The formulae for isotopic spin rotation are deducible from the
dichotomic nature of that variable and from the similarity of three
different twonucleon states, so there is no reason to abolish the
vivid geometrical terminology of "projections" and "rotations."
Sec. 32] ELECTRON SPIN 333
Let us now formulate the situation in a quantummechanical
fashion. For several nucleons, it is possible to define their resulting
isotopic spin operator.
T=JT< (32.18)
i
Its different components do not commute. But its square, T 2 , commutes
with one projection, say T*, which defines the resulting charge of the
given system. The Hamiltonian of the nuclear interactions (with
electromagnetic interactions neglected) commutes with both T 2 and
f z , just like the Hamiltonian for an electron in a central field
commutes (to a nonrelativistic approximation) with 2 and [i z . It
follows that T 2 and T* exist in nuclear states with a given energy.
In other words, nuclear states can be distinguished by their T 2 and
T* values. This ascribing of T 2 , T* to nuclear levels is approximate,
like the distinguishing of atomic levels by n, I, k, the difference being
that no account is taken of the magnetic properties of spin.
In heavy nuclei, electrostatic interactions are very important
because they increase in proportion to the square of the atomic
number. Nuclear interactions increase linearly with the number of
nucleons, as the mass defect of nuclei does. So in heavy nuclei both
types of interaction are of an equal order of magnitude and the neglect
of electrical interaction has no meaning. No definite isotopic spin
values can be attributed even approximately to the levels of heavy
nuclei.
The isotopic spin variables are very important in the classifying
of elementary particles.
Exercises
1) Write down the transformation of a x , a y , a z for any arbitrary rotation
in space and prove that the properties of the operators do not change.
The general expression for the transformation of vector component is
the following (see Sec. 9):
*ik = cos ( Z. x
where the coefficients a^ satisfy the conditions
f
where, according to the summation convention, n runs from 1 to 3.
2) Find the eigenvalues of the scalar product <jia of two electrons with
rallel an r
parallel and antiparallel spins.
We begin with the equation
a = a + af + 2a 1 o 2 .
o
But aj = o = a x + a y f a*= [see (32.11)]; the maximum projection of
a x + <* 2 is equal to zero for antiparallel spins, and is equal to unity for parallel
334 QUANTUM MECHANICS [Part III
spins. From this, (c^ f a 2 ) 2 in the first case and is equal to 12 = 2 in the
second case. This gives
T 3
o 1 o 2 =  ~  = T (
_____ ^
GI 02 =   _ (parallel spins) .
_5 b
3) Write down the eigenfunctions for a*, a y , and a z 
for <% = !. + = (J), for a,= i, *(_}).
for Oy  J, *  (J). for ,,  _ J, +
for.=i,* = (J), for,= l,4,
Thus, the eigenfunctions of all three noncommutative operators differ.
4) Express the scalar product j lt y a in terms of the resultant angular mo
mentum 7, /j, and ? 2 .
By definition j 2 j\ ~\ j\ 4 2y" 1< / 2 . Substituting here the squares of the
angular momenta, wo obtain
Ji J 2  y * *' 1} " jl (jl + l) ~ J2 (jz + 1)} *
Sec. 33. ManyElectron Systems
The Mendeleyev periodic law. Long before the atom became an
object of physical study its properties were investigated in chemistry.
And chemistry discovered and studied such properties of the atom
as were utterly alien to prequantum physics. In this category belongs,
first of all, valency or the chemical affinity of atoms. On the basis of
a vast quantity of experimental material accumulated in chemistry,
Mendeleyev constructed a generalizing and systematic periodic law.
This was a new law that allowed Mendeleyev to predict the existence
of many elements, which were discovered later. And what is more,
the basic chemical and many physical properties of these elements
were correctly predicted. At the present time, too, the Mendeleyev
law guides scientific investigation into the study of the periodic
structure of nuclear shells.*
The Fauli principle. The wave equation for a single particle is in
adequate for an explanation of Mendeleyev's periodic law. It is ne
* We have in mind the quantummechanical theory of nuclear shells,
and not the rather widespread speculative constructions which are mainly
based on the arithmetic relationships between the atomic numbers and atomic
weights of the elements.
Sec. 33] MANYELECTRON SYSTEMS 335
cessary to introduce a new principle concerning manyelectron
systems the socalled Pauli principle. We shall first of all formulate
it in such a way that it can be conveniently used to investigate the
electron shells of an atom, to wit, an atom cannot have more than one
electron with a given group of four quantum numbers : the principal
quantum number .n, the azimuthal quantum number I, the magnetic
quantum number ki and the spin quantum number k a . The spin
quantum number is a measure of the spin projection onto the same
axis onto which the orbital angular momentum is projected.
The Pauli principle is substantiated by relativistic quantum mechan
ics (see Sec. 38). Here we shall simply use it as a supplementary
principle of quantum mechanics.
The addition of angular momenta of two electrons with identical n
and /. We shall first of all show how the Pauli principle is applied in
adding the angular momenta of two electrons for which the principal
and azimuthal quantum numbers are the same. The accepted practice
is to say that these electrons belong to the same shell. Usually (though
not always) electrons in different shells possess quite different ener
gies a fact which justifies the classification by shells.
Let us take the simplest case when n = \. Then, in accordance with
the definition of n (31.33), 1 = 0. But, for I equal to zero, the magnetic
quantum number ki is also equal to zero. Hence, three quantum
numbers are the same for the electrons and, according to the Pauli
principle, the fourth number k a must differ. However, k a can only
have two values, + y and y , and each of its values can only have
one electron for given n, I, and &/. Thus, an atom can have only two
electrons with n~l. Their spins are antiparallel and therefore the
resultant spin S is equal to zero. The resultant orbital angular momen
tum L is also equal to zero.
Let us now take two electrons in the ^pstate, i.e., with I = I and with
the same principal quantum numbers. Either the magnetic or the spin
quantum numbers, or both, must differ. The ^electron can be in six
states, which we list writing the magnetic quantum number firsthand
the spin projection second:
It follows that two electrons can occupy any two different states of
the six. As is known, the number of combinations of six things, two
6x5
at a time, is equal to Cl= ^ = 15  These fifteen states differ by
the total orbital angular momentum L and the total spin 8, as well as
by their projections. The latter depend upon the choice of coordinate
axes and will interest us only insofar as they characterize the relative
directions of L and S.
336 QUANTUM MECHANICS [Part III
We shall first of all find those states which correspond to the greatest
projections of L and S, because they determine the possible eigen
values of L and S. In any case, of the fifteen states only those must be
taken, for which the total spin projections and orbital angular mo
mentum are nonnegative, since it is obvious that negative projections
cannot be greatest in relative value. The states with positive pro
jections number eight out of fifteen, and from these eight we take
only those which possess the greatest projections. We rewrite all eight
states :
AD: 2, 0; BD: 1, 0; CD: 0, 0; AB: 1, 1] AC: 0, 1; AE: 1, 0; BE: 0, 0;
AF: 0, 0.
The state with maximum orbital angularmomentum projection is
AD. Hence, a state exists for which the orbital angular momentum is
equal to two and the spin angular momentum is zero (here, and in
future, the angular momentum is characterized by the greatest pro
jection). The indicated state also yields projections 1, and 0, 0.
Such states are, for example, the BD and CD states; therefore they
need not be considered, since they do not define the vector sum. The
AB state has the maximum spin angular momentum. It follows that
a state exists for which the orbital and spin angular momenta are
equal to unity. Their projection can be 1, 0; 0, 1; and 0, 0. These are
AC, AE, and BE, which, like BD and CD, no longer interest us. There
remains one more state with projections 0, 0.
Thus, only three states are possible:
Composition of angular momenta for three electrons with identical
n and /. For three ^electrons we obtain the following seven states
with positive projections:
,; ACE:0, ; ABD:2, ; ABE: 1, ;
The maximum spin projection is 3/2 for a zero orbital angularmo
mentum projection. The maximum orbital angularmomentum pro
jection is 2 with a spin projection ^ . These two states, together with
their projections, are listed in order from ABC to ABF. ACD and
A BE remain, to which there correspond L 1 and S ^ . In all, we
have L = Q, S = 3/2; L = 2, S = ~] L = l, 8 = ~.
Normal coupling. States with different values of the total orbital
and spin angular momenta L and S, and with the same principal
Sec. 33] MANVKLECTKON SYSTEMS 337
quantum numbers for the electrons, differ in energy. This difference
occurs as a result of the electrostatic, and not magnetic, interaction
between electrons. In order to explain why the resultant orbital angu
lar momentum affects the interaction energy, we examine two ^elec
trons. The sum of their orbital angular momenta can yield two, unity
or zero. If two is obtained, then the angular dependence for the wave
functions of both electrons is the same (not only do the azimuthal
quantum numbers coincide, but also the magnetic quantum numbers,
that is why at least the spin projections must differ). We shall call the
wave functions of both electrons d^, t (i\) and fy l9 x (r 2 ), where r x and r 2
are the radius vectors of both electrons and the indices refer to the
quantum numbers I and k v
The interaction energy between electrons is approximately
because e \ fy l9 x (r x )! 2 and e d/ t , l (r 2 ) 2 represent the densities of the
charge distribution. The approximation consists in the fact that the
effect of the interaction on the wave functions and the socalled
"exchange" [see (33.32)] has not been taken into account.
If the resultant moment is unity, we correspondingly obtain the
other estimation:
Here the magnetic quantum numbers are equal to zero or
unity.
This integral is clearly different from the previous one. Thus, there
appears to be interaction between the orbital angular momenta when
they are to form the total angular momentum; this interaction does
not involve c 2 in the denominator, i.e., it is electrostatic in
character.
In multielectron atoms, Pauli's principle imposes definite conditions
on the choice of spatial wave functions for given spins. As an example,
let us consider the state with spin 3/2, which, as was just established,
is possible in a system with three ^electrons. In accordance with the
Pauli principle, three different spatial wave functions for the separate
electrons having ki = 1, and 1 correspond to this state. The corre
sponding electron densities coincide in space less than, for example,
in a state with magnetic quantum numbers 1, 1, 0, to which, according
to the Pauli principle, there must correspond spin projections ^ , ^ ,
  in order that all three pairs of &/, k a should be different. But the less
&
the electron wave functions coincide in space, the less the Coulomb
repulsion energy between the electrons, because the mean distance
between like charges is greater. For this reason, the state to which the
22  0060
338 QUANTUM MECHANICS [Part III
Pauli principle assigns the greatest possible spin possesses the least
repulsion energy.
There are three ^electrons in the ground state of nitrogen. As was
just indicated, the state with the least energy occurs when all three
spins are parallel. The next state, for which the orbital angular mo
mentum is equal to 2 and the 'spin is equal to  , lies approximately
2.2 ev higher, while the state with orbital angular momentum 1 and
spin T> lies 3.8 ev higher.
We can explain why a lesser energy corresponds to a greater result
ant orbital angular momentum. Wave functions, for which the
orbital angularmomentum projections differ only in sign, are closer
to each other than functions for which the angularmomentum pro
jections differ in absolute value. But those functions which correspond
to a closer spatial electron density distribution lead to a larger repul
sion energy, while angular momenta in opposite directions, when
summed, yield a lesser resultant angular momentum than the angular
momenta whose projections differ in magnitude also. Thus, the state
with the greatest spin possesses the least energy and, for a given spin,
it is the state with the greatest orbital angular momentum that has
the least energy (Hund's first rule).
This is the way the orbital and spin angular momenta are com
bined. In calculating the electrostatic energy only, the state of the
atom is defined by the absolute values of L and S. But a magnetic
interaction takes place between the resultant orbital angular mo
mentum and the resultant spin angular momentum of a system of
electrons, analogous to that of a separate electron (see Sec. 32). To a
first approximation, this interaction is described by the scalar product
A\LL x*, where JJLL and JJL, are the magnetic moments for the orbital
and spin motion of the system, and A is a factor of proportion
ality.
The scalar product of two angular momenta assumes as many
values as are possessed by the resultant angular momentum for a
given absolute value of the component angular momenta (see exercise
4, Sec. 32). This is clearly shown with the aid of a socalled vector
model: a triangle is constructed on the vectors L, S and J = L+S.
In accordance with the law of composition of angular momenta
(30.30), the side J can equal L + S, L + S 1, . . .,  L S \ . The
energy level of an atom with given values of L and 8 is split into as
many fine structure levels as can be assumed by J, i.e., 2 8 + 1 levels
if S is less than L, and 2 L + 1 levels if L is less than S.
The system of levels described here occurs with the socalled Russel
Saunders normal coupling (of orbital and spin angular momenta):
the energy states with different L and 8 differ considerably more than
the energy states with given L and 8 but different J. The group of
Sec. 33] MANYELECTRON SYSTEMS 339
energy levels differing only in total angular momentum J is termed a
multiplet.
In heavy elements, where the spinorbital interaction for separate
electrons is great, the spin of each electron in a shell is combined with
its orbital angular momentum to form a resultant angular momentum
j [see (32.15)]; only then do the angular momenta j of separate elec
trons combine. This may be accounted for by the fact that, the
relativistic effect of magneticmoment interactions is not small com
pared with the energy of electrostatic repulsion between electrons in
the inner regions of the atoms of heavy elements, where the electron
velocity is close to the velocity of light. The type of coupling which
occurs when the j of separate electrons are added, is termed jj
coupling, jj coupling also occurs between nuclear particles as a
result of the large spinorbital interaction characteristic of nuclear
forces.
The spcctroscopic notation for levels. In general form, the spectro
scopic notation for the resultant state of an atom is written thus :
2S + l]jg,u t
The main symbol is L, i.e., the letters S, P, Z), F, etc., depending
upon what L is equal to: 0, 1, 2, 3, .... As a left superscript we put
2 S + 1. As a right subscript we put J, i.e., the vector sum of L and
S from the number of the finestructure components. Finally, the
right superscripts denote an odd (u) or even (g) state, respec
tively.
For example, the ground state of a nitrogen atom has L = 0, $ = 3/2
and is formed by three ^electrons. Hence, its spectroscopic designation
is 4 $?/ 8 because the total angular momentum can only equal the
spin angular momentum (L = 0), and JT*Z = 3 is an odd number.
The notations for the next two states of nitrogen are
2 D" and 2 P",
or, if the multiplet splitting is taken into account, then
or 2 D and 2 Pj or
depending upon the resultant angular momentum J.
If the ground state of the atom has L and S not equal to zero, the
resultant angular momentum is determined by Hund's second (empir
ical) rule : when there are less than half the possible number of elec
trons in a shell, the least energy corresponds to a multiplet level for
which J=\L S , and to that of J = L + S when there is more
than half the possible number. Since the electron angularmomentum
Z can have 21 + 1 projections, and there are two values k a for each
22'
340 QUANTUM MECHANICS [Part III
projection &/, then there can be in all 2 (2 l + l) electrons in a shell
with given values of I and n. The total number of electrons in an
atom with a given principal quantum number n is
Hl
l)~~=2n*. (33.1)
The electron configuration corresponding to the least energy occurs
in the ground state. It is determined by Hund's first and second
rules.
The dependence of energy on the azimuthal quantum number.
Before we can go over to a description of the Mendeleyev periodic
system we have to remark on the dependence of the energy of an elec
tron on the azimuthal quantum number. The energy of an electron
in all atoms, except the hydrogen atom, depends upon / as well as
upon n. For large / the electron is situated comparatively far away from
the nucleus; in other words, it is more weakly bound to the nucleus
than for small L For a given n, the energy of an electron is greater, the
larger I. When the field greatly differs from a Coulomb field, the de
pendence of energy upon I is so strong that an increase in the princi
pal quantum number n, with a simultaneous decrease in I, leads to a
smaller energy increase than the increase of I for a given n. In other
words, the state with quantum numbers n + l, can have a lower
energy than the state with quantum numbers n, I. This will become
clear in the later examples.
Filling the first shells. As was mentioned, the shell with n = l is
filled by two electrons in the Isstate (the 1 in front denotes the quan
tity n). Hydrogen has one electron in this shell and helium has two.
The helium shell is completely filled and has a 1 S$ state. The electron
configuration for the ground state of a helium atom is so stable that
if any other atom approaches close to it the total energy can only
increase, so that repulsion forces are produced. Helium is completely
inert chemically. The forces between helium atoms are small as a
result of the symmetry and stability of their electron shells. Therefore,
helium gas is liquified at an extremely low temperature.*
After helium, the shell structure with n = 2 begins. The first electron
of this shell, i.e., a 2 ^electron, appears in lithium. The two inner
Iselectrons occurring in the helium configuration strongly screen
the nuclear charge and, consequently, the outer electron is weakly
* The condensation of helium into a liquid at low temperatures is due
to the socalled Van der Waals forces, which arise out of the mutual electro
static polarization of approaching atoms. These forces act at larger distances
than the forces of chemical affinity, and are very small compared with them.
Sec. 33] MANYELECTRON SYSTEMS 341
bound. Such is the alkalimetal electron configuration in the case
of lithium, and analogous electron configurations subsequently result
each time (Na, K, Rb, Cs) from the addition of an ^electron to a nucleus
surrounded by a noblegas electronic cloud. The next 2 ^electron
has an energy which is comparatively close to the energy of a
2 ^electron : the energy of the electron is still weakly dependent on the
azimuthal quantum number since the field is approximately Coulomb.
A large energy is needed for an electron to go from a 1 sshell to a
2 5 or a 2 pshell, while a small energy is needed for the transition
from a 2 s to a 2 ^shell. For this reason, the beryllium electronic
configuration, having two 2 selectroiis, is not very stable with respect
to an electron transition to the 2 pshell. In other words, filling the
2 sshell does not give the electron configuration of a noble gas. Indeed,
as we know, beryllium is a metal.
After beryllium, the 2 pshell fills up, and is completely filled for the
noble gas neon. Neon follows fluorine, which requires one electron for
the shell to be filled. The energy required for an electron to be added
to the fluorine 2 pshell, to fill the shell of neon, is large. This explains
the chemical activity of fluorine and the other halogens, which are
similarly situated with respect to the noble gases.
There can be eight electrons in a shell for which n = 2. This is the
first group of the Mendeleyev system. The shell with n = 3 is then
filled, though initially only the first two subshclls : 3 s and 3 p. The
elements of the second group have an outer electronshell structure
similar to the elements of the first group. The chemical properties of
atoms are basically determined by the outer shells. This explains the
similarity of chemical properties, on the basis of which Mendeleyev
formulated his law. Argon has a filled shell, i.e., still another group
of eight elements is completed. The noblegas configuration is obtained
for argon because the 3 ^state, on the one hand, and the 3 d and 4 s
states, on the other hand, differ considerably in energy.
By considering the possible states of shells which , to be filled, lack
less than half the possible number of electrons, we can consider that
unfilled states behave like electrons. For example, if there are two of
the six electrons wanting in a 2 pshell, then we can combine the
states of the two "holes," similar to the way that the states of two
2 pelectrons were combined at the beginning of this section. In doing
so, correct results are always obtained, provided that Hund's second
rule is used in finding the total angular momentum J of the ground
state, i.e., that we take J = L + S. It is easy to see that four electrons
in the shell are equivalent to two holes by applying the Pauli principle
first to an electron and then to a "hole," (see exercise 2).
Let us now give, in one table, the scheme for building up the first
eighteen places in the periodic system of elements ; this table shows
the number of electrons having given quantum numbers.
342
QUANTUM MECHANICS
[Part III
Element
n = 1,
J 
n = 2,
/ 0
n =2,
/ = 1
n = 3,
J =
n = 3,
I = 1
Ground
state
H
1
8f h
He
2
* s o
Li
2
1
*S! /t
Bo
2
2
l s$
B
2
2
1
n
C
2
2
N *%
2
2
3
4 ^ M / a
()
2
2
4
3 pf
F
2
2
5
2JP "/I
No
2
2
6
l s%*
Na
2
2
6
1
2 sf /t
Mg
2
2
(5
2
l s{
Al
2
2
6
2
1
2jP "/ a
Si
2
2
6
2
2
3jp o
P
2
2
6
2
3
* S "lt
S
2
2
f
2
4
*P%
C'l
2
o
2
5
2p"
/a
AT
2
2
(i
2
(3
*^0
The filling order after the 3/ishell. After argon, the 4sshell begins
to fill instead of the 3rfshell. The new group begins with the alkali
metal, potassium. The sum n + l is the same for the 3p and 46 l shells
and is equal to 4, while it is already greater by unity in the 3dshell.
The 4^}shell is filled after the 3rfshell, with the same value of the
sum n +Z = 5, and then the 5sshell. It is seen that this rule is observed
later on, too; the filling of the shells with the same sum n\l
proceeds in order of increasing ?i. But there are certain deviations
from this rule during the filling of the d and /shells.
In the shells with n=l, 2, 3, there are altogether 2.1 2 + 2.2 2 +
+ 2.3 2 = 2 + 8 + 18 = 28 electrons. There are a further eight electrons
in the 4s and 4pstates, and another two electrons in the 5sstate.
The 55state is followed by electrons with n + Z 6, where we begin
with the least n, i.e., witli 4d. There are 2 (4 + 1) 10 more of these
electrons. The 4delectrons are followed by 5>electrons, of which
there are six, and then by the same simple rule we get the 6sstate.
Hareearth elements. The next value of n + l = 7, the least being
7i = 4. Hence, beginning with the 57th place (in actuality, with the
Sec. 33] MANYELECTRON SYSTEMS 343
58th place) the 4/shell can begin to fill acquiring at once two 4/
electrons. This shell is already inside the atom as a result of the
form of the potential distribution within the atom. The screening
of the nuclear charge by atomic electrons leads, at large distances
away from the nucleus, to the potential decreasing like . instead
1 r
of (see Sec. 44). If we combine the potential energy of an electron,
calculated with allowance for screening with the centrifugal energy,
it turns out that the d and /states possess a minimum resultant
effective potential energy deep inside the atom (see Sec. 5, footnote
to p. 45). Indeed, the centrifugal energy is greater than the potential
energy both close to the nucleus (with allowance made for screening)
as well as far away from the nucleus. Therefore, the effective poten
tial energy UM is positive for large r as well as for small r. In other
words, for the d and /states, the UM curve goes higher for large r
than for the s and ^states, and it turns out that the effective poten
tial well for d and /electrons is situated closer to the nucleus than
to the boundaries of the s and pelectron shells. Thus, the d and
/shells are, as it were, filled inside the atom. But the chemical prop
erties of atoms depend mainly upon the outer electrons which,
in filling the 4/shell, change very little. This is how the group of
2 (2*3 + 1) =14 chemically similar elements, termed rareearth
elements, originates.
It should be pointed out that the d and /shells arc not filled
successively as a result of "competition" with outer shells : for example,
there are throe delectroiis and two ^electrons in F 23 , five rfelectroiis
and one selectron in the next element O 24 , while Mn 25 also has five
dolectrons but two .^electrons.
The statistical theory of the atom, which will be set out in Sec. 44,
permits us, in rough outline, to find the potential distribution inside
an atom. It becomes possible, from this distribution, to predict
rather accurately the places in. the periodic system where elements
with I 2 and 3 appear.
The 5 /shell fills up (beginning with thorium) in a whole group
of elements similar to the rare earths. A large part of this group
consists of the artificially produced transuranium elements.
The wave equation of a twoelectron system. We shall now formulate
Pauli's principle using wave functions. The simplest way to do this
is to consider a twoelectron system. The wave equation for two
electrons may be written thus:
=  ~a A I A * = U < f i' r *> * = ** (332)
Here AJ and A 2 are the Laplacian operators with respect to variables
of the first and second electrons, U(r ly r 2 ) is the potential energy
344 QUANTUM MECHANICS [P&Ft III
of their interaction with the external field and with each other.
For example, in a helium atom
9/>2 0/>2 ,,2
U (r l5 r.)   4'   7 +  ' . (33.3)
r l r 2 I F l ~ F 2 I
The wave function depends upon the spatial and spin variables
of both particles:
fcOO^ir^). (33.4)
The interaction of spin magnetic moment with orbital motion
is weak. Therefore, to a first approximation, the spinorbital inter
action can be neglected in the potential energy operator. This cor
responds to U (FH r 2 ) in equation (33.3). Tf the effect of spin motion
upon orbital motion is small then the probability of a certain value
of spin and coordinate is equal to the product of the probabilities
of both values, and the probability amplitude < also divides into
a product of amplitudes
(D (r l5 a l ; r 2 , a 2 )  T (r 1? r 2 ) x (a,, cr 2 ). (33.5)
The probability amplitude of orbital motion satisfies equation
(33.2), provided it does not involve the spin operator. But even
when the system is placed in an external homogeneous magnetic
field II, the following operator is added to f jfr
where it is taken that the zaxis coincides with the direction of the
field (the minus sign is replaced by a plus sign because the electronic
charge is negative). The action of the operator S zl \ S Z2 on the
spin function simply gives the total spin projection of the system.
For this reason, in the presence of an external homogeneous magnetic
field, U mdK is replaced by a number which is added to the total energy
of the system.
The symmetry of the & operator with respect to particle inter
change. When examining the jfe operator in (33.2) we see that it is
completely symmetrical with respect to a coordinate interchange
for both particles, i.e., it does not change its form if the first electron
is called the second, and the second electron the first:
JT (r 1? ^ ; r 2 , o 2 )  3? (r a , a 2 ; r^ aj. (33.7)
But equation (33.2) is linear. Therefore, if the form does not change
due to operation (33.7), then the wave function can only be multiplied
by some constant number P:
$ (r lf aj; P 8 , cr 8 )  PO (r 8J a 2 ; r lf a t ). (33.8)
SeC. 33] MANYKLKCTRON SYSTEMS 345
Because r x , G! and r 2 , <r 2 are involved in the saino way in all the
equations, we can interchange them in (33.8) obtaining
O (r 2 , <j 2 ; r 1? aj  PO (r x , cv, r 2 , cr 2 ). (33.9)
Substituting (33.9) in (33.8), we shall have
or
pa=l, P=l. (33.10)
In this comparatively simple case, when there are only two par
ticles, the transformation is similar to the symmetry transformation
for a wave function under reflection [see (31.38)].
The commutative operator for coordinates and spin variables.
We can define a coordinate commutative operator for electrons
P T such that
r 1 ). (33.11)
If the wave equation is symmetrical with respect to interchange
of r x and r 2 (without interchanging the spin variables v l and <r 2 )
then, by repeating the foregoing argument, we see that the eigen
values of Pr for two electrons are equal to 1.
An analogous operator is also defined for spin
where the eigenvalues of P a are likewise equal to h 1.
We denote the set of orbital quantum numbers of the first electron
by the letter n (in place of n l9 l, k/ t ), and those of the second elec
tron by the letter n 2 . Then the orbital wave function X F is written in
more detail as
It follows from requirement (33.10) that
P r x F (14, r x ; n a , r a )  T (n l9 r 2 ; n 2 , r x )  T (n l9 r x ; n a , r a ) . (33.13)
The function in (33.13) with an upper sign is termed symmetric;
with a lower sign, antisymmetric.
The wave function of a twoelectron system. Introducing, in ad
dition, the spin quantum numbers k ai and & 02 , which determine
the form of the spin wave functions (see exercise 3, Sec. 32), we write
the total wave function of a twoelectron system as
340 QUANTUM MECHANICS [Part 111
The total permutation of spin and spatial coordinates in this
function occurs as a result of the action of the P operator, which is
P=PrP a . (33.14)
Operating with (33.14) on the function O, we have
PO (WJL, fc 0l , T 19 d, ; n 2 , k<, 2 , r a> a 2 )  O (w x , A 0l , r 2 , cr 2 : w 2 , fr 0a , r l5 orj . (33.15)
According to (33.10) this function is also either symmetric or
antisymmetric. But it can now be seen immediately that only the
antisymmetric function satisfies the Pauli principle. Indeed, let
the states of both electrons be identical, i.e., n^^n^ k ai k at . Then,
if the function O is antisymmetric, we obtain
PcD ( Wl , 4 0l> r x , (jj; n v fc 0l , r 2 , d 2 ) = O (n l9 fc 0l , r a ^i ? h> ^ r^ a L ) =
= O (n l9 k ai9 r l5 <7i ; n x , fc 0l , r a , cr 2 )
=^ O (n l5 fe 0l , r 1? ^ ; n l9 k ai , r 2 , cr 2 ) .
(33.16)
"By definition, the operator P interchanges only the variables r
and or, and by no means the quantum numbers ?i, k a . The first
equation of (33.16) denotes the result of a P operation, the second
takes into account the antisymmetry of the wave function, while
the third is obtained from the first by permutation of all four
arguments relating to the electrons. The possibility of such a per
mutation for any function is obvious, since it does not matter which
particle is considered first and which second when writing down
the wave function; the interchange of the four values n x , & 0l , r l c^
and n t , k 0l , r 2 , <7 2 m tnc ^ ast equality of (33.16) simply does not
denote anything: it is immaterial which arguments are written
first those relating to the first electron, i.e., n^ k ^ r t , <r t , or those
relating to the second electron w , k 0l , r 2 , o 2 . Hence, the function
* (MI* ^<v r i a i> n i* ^i r 2> ^2) is equal to itself with the sign reversed,
i.e., it becomes zero identically.
This property is possessed only by an antisymmetric function
and not by a symmetric function ; the latter would become identically
equal to itself. But if the antisymmetric wave function of two elec
trons occurring in identical states is identically equal to zero, the
probability amplitude of this state of a system of two electrons
is equal to zero for any values of the variables r t , r 2 , c^, a 2 . Only
an antisymmetric function is compatible with the Pauli principle.
The same applies to the wave function for a manyelectron system :
it is antisymmetric with respect to a simultaneous permutation of
spatial and spin variables for any electron pair. This is the generalized
formulation of the Pauli principle.
Sec. 33] MANYELECTRON SYSTEMS 347
Particles with halfintegral spin. Experiment shows that all ele
mentary particles with half integral spin obey the Pauli principle:
protons, neutrons, electrons, and positrons. Complex particles,
consisting of an even number of elementary particles with halfintegral
spin, have a symmetric wave function because, for a complete inter
change of all variables relating to such a complex particle, we must
make an even number of permutations of the elementary particle
variables it consists of. But by changing the sign an even number
of times we do not change it at all. For this reason, nuclei with even
atomic weights (for example, D 2 , He 4 , N 14 , O 16 , etc.) and, therefore,
having symmetric wave functions arc not subject as units to the
Pauli principle, while He 3 , Li 7 , etc., have an antisymmetric wave
function, that is to say, they are subject to the Pauli principle.
Elementary particles not subject to the Pauli principle. Light quanta
do not obey the Pauli principle since there can be an unlimited
number of quanta in a state with a given wave vector k and given
polarization. All particles with integral spin possess a wave function
which is symmetric with respect to a complete permutation of the
variables relating to any pair of particles.
The Pauli principle and the limiting transition to classical theory.
The Pauli principle enables us to understand why the wave prop
erties of light quanta are conserved in the limiting transition to
classical theory, while the wave properties of electrons are
not.
We shall consider quanta in definite states, i.e., having a certain
polarization and wave vector. The number of such quanta can be
infinitely large, since quanta are not subject to the Pauli principle.
We note that this was not introduced as a supplementary hypothesis
concerning the properties of light quanta, but was directly obtained
in Sec. 27 in the quantization of electromagnetic field equations:
the number of quanta in a state N kt is the quantum number for
the corresponding oscillator. If this quantum number is large then
the motion of the oscillator becomes classical, and, as we know,
its oscillation amplitude is proportional to the amplitude of a field
with a given polarization and wave vector. Thus, the limiting transi
tion yields a classical wave pattern.
In accordance with the Pauli principle, there cannot be more
than one electron in each state. Therefore, the probabilityamplitude
absolute values are always limited by the normalization to unity
and, consequently, do not pass to wave amplitudes which can be
defined classically. f
The ortho and parastates of two electrons. Let us now return
to the case for which the wave function can be represented in the
form (33.5). Since the whole product is antisymmetric, one of its
factors must be symmetric and the other antisymmetric. This simple
result refers only to the twoelectron problem.
348 QUANTUM MECHANICS [Part III
Let us consider the wave function for two electrons. Since the
spin of each electron is equal to (in atomic units), the resultant
2i
spin can only be equal to zero or unity. Both these states of a system
of two electrons have special names. The state with spin unity is
termed the orthostate, while that with spin equal to zero is called
the parastate.
As has already been said, the magnetic interaction of spins is
small. If we can neglect it, then it is easy to write down the spin
wave functions for the ortho and parastates. Let x (^a^ <*i) be a
function of the spin variable for the first particle c^, assuming, as
we know, only two values a t = 1 and (j 1 = 2. & CTI denotes the eigen
value of the spin projection. Depending upon whether fc 0l is equal
to  r or , , the function ^ has the form shown in exercise 3, Sec. 32.
Z Zi
Without assigning a definite form to x we write down the spin
wave function for two particles which do not have a spin magnetic
interaction :
CT I) X (*o a , * a ) < (33.17)
i.e., the probability amplitude for both particles breaks up into the
product of the amplitudes for each particle separately. However,
it must be taken into account that this function must be either
symmetric or antisymmetric. If & ai A;^, then (33.17) is symmetric
by itself:
x!> ^ixf* ><* 2 >
(33.18)
/ I \ / i \
x( 2 *i)x( 2 ' CT2 /'
For k ai ^k<, z we must form either a symmetric or an antisymmetric
combination of ^:
frvm  , 1 9 [x ( 1 > CT i) X (~ \ > ^a) I X ( 2 ^2) X ( 2 ' ^j] ' (33.19)
(33.20)
 is introduced for normalization.
V2
The magnitude of the spin projection, i.e., 0, 1 or 1, depends
upon the choice of the zaxis. But the symmetry or antisymmetry
of a wave function is an internal property and cannot depend upon
the choice of coordinate axes. Therefore, the state (33.19) must be
regarded as one state together with (33.18), if we judge by the total
spin value. They are distinguished by the spin projections, and the
total number of these states is three, as is required for a total spin
equal to 1. The upper line of (33.18), as is evident, corresponds to
Sec. 33] MANYELECTRON SYSTEMS 349
a total projection 1, the lower line to a projection 1, and (33.19)
to a projection 0. (33.20) corresponds to a total spin of zero. In the
accepted terminology, the state with unity spin is to be regarded
as the orthostate while that with zero spin, the parastate.
This definition of ortho and parastates from the symmetry of
the spin wave function also holds for particles with spin other than   .
But the resultant spin in the ortho and parastates turns out, in
this case, to be ambiguously related to the symmetry of the function.
The example of deuterons with spin 1 will be examined in
Sec. 41.
Ortho and parastates of helium. The two electrons in the helium
atom can occur either in the orthostate or the parastate. In the
first case, the atom has spin unity, in the second case, zero. The
symmetric and antisymmetric spin functions are the eigenfunctions
of the spincommutation operator P a : when operated upon by
the operator P they give fc l, i.e., the eigenvalues of P . To the
approximation (33.2)(33.3), the Hamiltonian* is commutative
with P , so that P a is an integral of motion. Therefore, transitions
between the ortho and parastates, during which the total spin
is not conserved, are far less probable than transitions with conser
vation of spin.
Only when spinorbital interaction is taken into account, when
the wave function cannot be expressed as a product of the form
(33.5), is P a not an independent integral of motion. But the cor
responding terms in the Hamiltonian,** which describe the spin
orbit interaction, are inversely proportional to c 2 . To this approxi
mation, it is only the total permutation operator of the spin and
spatial variables for both electrons P that is an integral of motion,
because the total wave function of the two electrons is always
antisymmetric in accordance with the Pauli principle.
The eigenfunction of a hydrogen molecule in a zero approximation.
Concluding this section we shall consider the quantum mechanical
explanation for the homopolar chemical bond. Such a bond occurs,
for example, between the two atoms in a hydrogen molecule. It
was first considered by Heitler and London.
We assume that the atoms are independent in the zero approx
imation. Each electron is situated close to its own nucleus. We
shall denote the nuclei by the letters a and 6, and the electrons by
the numbers 1 and 2. In the initial approximation, the interaction
* The Hamiltonian means the Hamiltonian operator.
** See A.I. Akhiezer and V. B. Berestetsky, Quantum Electrodynamics,
GTTI, 1953, equation (37.10). [English translation by Consultants Bureau,
Inc. New York, N. Y., 1957.]
350 QUANTUM MECHANICS [Part III
between atoms is not taken into account. But this does not mean
that the wave functions of two electrons can be taken in the form
because this function is neither symmetric nor antisymmetric with
respect to the interchange of the electron coordinates. Neglecting
the spinorbital interaction, we must write the spatial wave function
in one of the two forms:
or
t ^(r a ^(r b ,)+^(r a2 )^(r hl ), (33.21)
1 = * (r ai ) * (r^)  4/ (r at ) ty (r bl ) , (33.22)
assuming that the total wave function O is obtained by multiplication
of the spatial function by the spin function of opposite symmetry.
In this form the wave functions arc compatible with Pauli's principle.
We write r ai and r/, 2 in scalar form because the wave functions
of a hydrogen atom in the ground state do not depend upon the
angle.
The wave equation for a hydrogen molecule has the following
form :
h* . __ _h*_ . _ _h*_ . __ h* . _ e*_ _ e* _
2m p 2rn p 2m 2m T^ ^b t
(33.23)
The first two terms describe the motion of the nuclei of the molecule.
They involve the mass of the proton m p in the denominator, and are
therefore exceedingly small compared with the terms describing
the motion of the electrons. Physically, this means that the nuclei
move considerably slower than the electrons, so that we can find
the electron wave function for a fixed distance between the nuclei.
Then $ is a function of the distance between the nuclei. If this func
tion has a minimum, corresponding to a stable equilibrium for a
given electronic state, then it becomes possible for atoms to form
a molecule. We shall not in future write the terms corresponding to
the nuclear kinetic energy: they must be taken into account when
we consider the vibrational, rotational or translational motion of
molecules, though the very position of stable equilibrium, which
is determined by the electronic motion, can be found without allow
r h * A j W A
ance for = A* and ^ A& .
2m p 2m p
The terms of the Hamiltonian appearing in the first line of (33.23)
refer to separated atoms. We shall call this part (without ^~ A a
and ^Abj J^ , where the index indicates the degree of
the approximation. The second line involves terms due to atomic
Sec. 33] MANYELECTRON SYSTEMS 351
interactions: the attraction of electrons to * 'alien" nuclei, and the
Coulomb repulsion between electrons and nuclei. We shall call this
part c/i^ and consider it a perturbation : this is true, strictly speaking,
only in a qualitative sense.
Perturbation method. Using the notation 3Jf Q and J^, the wave
equation is written as
^ T  ( t ^ + Jf^) X F = *F. (33.24)
The energy eigenvalue can be expanded as the sum of the energies
of the noiiinteracting hydrogen atoms [from (31.34)] and the inter
action energy of the atoms:
f = ? + ?i. (33.25)
We consider the operator ^f l and the energy as corrections. Ac
cordingly, we separate the wave function into a zero approximation
function given by one of the expressions (33.21) or (33.22), and a
correction T 1 :
TTO + T!. (33.26)
We shall neglect the quantities 3F ^ l and ff^ X F X because, in our
approximation, they can be considered as being of the second order
of magnitude.
Substituting (33. 25) and (33.26) in (33.24) and omitting these small
terms, we obtain
(33.27)
But from the definition of X F we have
^o% = ?<>%> (33.28)
since ' is the zero approximation energy and YQ is an eigenfunction
of & .
We multiply the remaining terms by TO and integrate over the
volume dV 1 dV 2 of both electrons*:
= ^ JT 2 d VidV* + tfojV^dVidVz. (33.29)
We can transform the first term on the left making use of the
fact that J^ Q is an Hermitian operator ( X F is a real wave function) :
oY 1 dF 1 iF, > (33.30)
* For simplicity, we consider here the real Hamiltonian and the real wave
functions; in the more general case, we must multiply the left by *FJ.
3/52 QUANTUM MECHANICS [Part III
so that it cancels with the last term ou the left, in accordance with
(33.28). The same could have been proved by using a definite form
for J#" in the given case.
Finally, from (33.20) we obtain
(33.31)
The denominator of this expression is the square of the normaliza
tion factor of the function X F , which square appears because the
expressions (33.21) and (33.22) are not normalized to unity. Chang
ing to a normalized wave function,
we represent the energy correction to a first approximation in the
form
,? 1= = >;JVF;dMF 2 . (33.31 a)
This expression is equal to the average of the perturbation energy
over nonperturbed motion [see (30.24)].
It can be seen from the very deduction of equation (33.32), in
which only the general Herinitian property for operators is used,
that the result is of a general nature.
Hound stale of hydrogen molecule. From (33.21) or (33.22), we can
substitute either a symmetric or an antisymmetric coordinate func
tion in the formula for the energy of a hydrogen molecule. Evaluating
the integral shows that l (r it b) has a minimum for the symmetric
form of the spatial wave function only. The depth of this minimum
corresponds approximately to the binding energy of a hydrogen
molecule. We cannot expect very good agreement with experiment
here, since the approximation used was more qualitative than quanti
tative in nature.
In calculating the energy, the following integral is of essential
importance
* 2 fy (r*i) * (r*,) ~ * (r,) } (r bl ) d V, d V, , (33.32)
J ^12
and is termed the exchange integral. It cannot be correlated with
any classical quantity because it involves only probability ampli
tudes and not densities.
We note that the antisymmetric spatial function (33.22) possesses
a nodal surface between the nuclei because it changes sign when
the places of the nuclei a uiid 6 are exchanged. A symmetric func
See. 34] TITK QUANTUM THEORY OF R UHATIOX 353
tion does not have nodes and therefore corresponds to a smaller
total energy, i.e., a more stable state. This function is multiplied
by an antisymmetric spin function so that a stable molecular state
has a zero resultant spin. Thus, the homopolar bond of two hydrogen
atoms forming a molecule is related to a "saturation, 13 as it were,
of spins. There is "no longer a stable equilibrium position for a third
hydrogen atom near a hydrogen molecule.
The tendency towards spin saturation of electron pairs is pro
nounced in homopolar bonds.
Exercises
1) Find the possible states of a system of two ^electrons with the same
principal quantum numbers.
Each electron can occur in ten states:
A  9 7? i r* c\ /) i // 9 . JF . o fj i
' ~2~ ' ' 2 ~ 2 2~ ' ' ~ 2 ' "'  2 ' ~~ 2' '
The states with positive projections of spin and orbital moment are
AB: 3,1; AC: 2,1; AD: 1,1; AE: 0,1; AFi 4,0; Alii 3,0; AH: 2,0;
AI: 1,0; AJ: 0,0; 7?C7: 1,1; BD\ 0,1; /*F: 2,0; /*<7: 2,0; /*// : 1,0; 7*7: 0,0;
CFi 2,0; C'tf: 1,0; CH : 0,0; 7>7< T : 1,0; DO: 0,0; #7< T : 0,0.
Choosing the states with maximum angularmomentum projections, wo
obtain three resultant states with zero spin :
1 S, 1 D, 1 G or 'AS*, ^, 1 G$,
and two states with unity spin:
3 P, 3 F or 3 Pf , 3 Pf , 3 Pg and 3 i^f , 3 J^, 3 Ff .
2) Show that, in a system of four pelectrons with the same principal num
bers, the states are the same as in a system of two ^electrons ; in other words,
that two electrons have the same states as two '"holes."
Sec. 34. The Quantum Theory of Radiation
In this section we shall find the probability of an excited atom
emitting a light quantum in unit time, and we shall compare the
probabilities of such radiation transitions as correspond to various
changes in the atomic states. But lirst we must deduce a general
formula for the probability of quantum transitions (this formula
will also be used in Sec. 37).
Transitions between states with the same energy. Let us suppose
that a system has two states corresponding to the same energy but
different in some other respect. For examj>le, in this section we
23  0060
354 QUANTUM MECHANICS [Part III
will consider an excited atom having energy excess ff : above
the ground state. This atom is capable of emitting a light quantum
with energy A CD =ffi <? , in which case the atom will go to the
ground state. Strictly speaking, there is only one state consisting
of an atom and electromagnetic field with a certain energy (f x (if
there are no other quanta in the field). The energy of such a system
is rigorously determined, though the state is not defined in more
detail.
Also possible is the following approach to the problem. Let an
atom at the initial instant of time be in an excited state but capable
of the spontaneous emission of a quantum. Then the energy of the
atom is no longer specified with full rigour, but lies in some narrow
interval A (^, where A $ ~ ~r an( l ^ ^ ^ ie mean lifetime of the
atom in the excited state before radiation [see (28.15)]. If the mean
lifetime of the atom in the excited state A is such that ^ is con
A
siderably less than the energy level spacing of the atom, then the
energy uncertainty can be neglected to a first approximation,
assuming that the atom initially occurred in a state with an accurate
energy value S^\ it is also necessary to calculate the probability
that, in a certain interval of time , the atom will go to the ground
state, and a quantum with energy h CD =ff : <^ will appear in the
electromagnetic field.
The reason for the transition is interaction with the electromagnetic
field. Here the lifetime of the atom in the excited state A is so great
that &< > <^6\ <? . For this reason, the interaction of the atom
with the electromagnetic field can be interpreted as a small pertur
bation superimposed on the excited atom with energy .
The same type of problem concerning the transition probability
due to a perturbation can also be formulated for other transitions.
For example, if the total excitation energy of an atom is greater
than its ionization energy, it is possible for an electron to be emitted
from the atom without radiation. In this case the excited state of
the atom and the ion + electron state belong to the same energy.
Each of them separately does not have a strictly defined energy.
Transition probability. A radiation transition with the emission
of a quantum is caused by interaction between an atom and an
electromagnetic field. We shall suppose for the time being that this
interaction is "switched off"; then the energy of the atom and field
separately becomes an exact integral of motion. We shall call its
eigenvalue in the initial state S^. Then, if the interaction is "turned
on," a finite probability exists of the system making a transition
to some state which, energetically, is very close to & 19 but otherwise
very different from the initial state; for example, the atomlwas
excited in the initial state and there were no quanta in the field,
SeC. 34] THE QUANTUM THEORY OF RADIATION 355
while iii the final state the atom went to the ground state and a
quantum appeared in the field.
Let us divide the Hamiltonian of the system into two terms:
je= <&w +J&M , where jfW corresponds to the separated atom and
field while ^ (1) describes the interaction. We then deduce a general
formula for the transition probability, and apply it to a radiation.
We shall therefore call 3t?w> the Hamiltonian of the unperturbed
system, and regard 3?) as a small perturbation causing the transition.
The eigcnfunctions and eigenvalues of the operator 30*^ are deter
mined from the equation
Allowing for perturbation, the wave function satisfies the equation
. . (34.2)
Considering that Jff^ is a small perturbation, we represent the wave
function in the form
(34.3)
the "product" &*& ij/ 1 * will be neglected as being of the second order.
Then, for tp (1) we obtain the nonhomogeneous equation
> (34.4)
We shall look for i]/ 1 * in the form of an eigenfunction expansion
of the operator t]/):
* (1) =27 c W*2 > , (34.5)
m
each of the functions 3$ satisfying the homogeneous equation (34.1).
Substituting the series (34.5) in the nonhomogeneous equation and
using the indicated property of the function fy\ we arrive at the
following equality:
W  (34>6)
The coefficients c m can be determined therefrom by taking advantage
of the orthogonality property of the eigenfunctions fy$ (30.6). For
this it is necessary to multiply both sides of (34.6) by ^n 0) * and integrate
over a volume. Then only the term   =~ remains on the left,
while on the right a certain integral is obtained which is characteristic
of the perturbation method set out here:
23'
356 QUANTUM MECHANICS [Part III
< 34  7 )
In order to integrate this equation we must determine the time
dependence of the righthand side. It involves wave functions [that
satisfy equation (34.1)] together with their time factor in the form
(24.21). It is assumed that the operator J^ (1 > does not depend ex
plicitly on time. Then
F = P""' ~ ' * " tf$*jf WF, (34.8)
where the integral multiplied by the exponent does not depend
upon time. We have supposed that at the initial instant of time
the system was in a state with energy ff\ in other words, that
Iq (0)  = 1, c W7 ti (0) 0. Therefore, equation (34.7) is integrated thus:
(34.9)
or, once again introducing the exponential factor under the integral
sign, i.e., reverting to the functions ^, ( , 0) * , ^i 0) we obtain
Cn (t)= l ~~* h U^tfW^dV. (34.10)
Consequently, the probability that at the instant of time t the
system will be in a state with wave function d>, ( , 0) is, from (30.13),
equal to
Matrix elements. We shall be concerned with the expression (34.11)
somewhat later. First of all let us introduce a system of notation
which, in general, is very convenient in quantum mechanics. The
integral (34.7) for any pair of eigenfunctions d w , AJ and for an arbitrar} 7
operator X is denoted thus:
X^MdF, (34.12)
Sec. 34] THE QUANTUM THEORY OF RADIATION 357
where the integration is performed over all the independent variables
involved in the Hamiltonian C/C.
The quantities Xk form a square table so that the index n will
designate the rows, and k the columns:
22
............. (34.13)
Such a table is termed a matrix in mathematics, while the separate
quantities Xfc are called matrix elements. The righthand side of
(34.7) contains the matrix element 3f^.
We note an important property of the matrix elements of Her mitian
operators. In accordance with the Hermitian condition (30.3)
, (34.14)
where the conjugate sign* on the right refers to the whole integral.
Proceeding now from the definition (34.12), we write:
X^Xfe,,. (34.15)
A matrix whose elements satisfy equation (34.15) is termed Hermitian.
The relationship between matrix elements of different quantities.
Let us take the matrix elements of both sides of the operator equalities
(30.35) and (30.37), and put the time derivative before the integral:
~dt Xnk ^ P" k ' (34.16)
The time dependence of the matrix elements was found in equation
(34.8), namely
Therefore,
~ Xnk = I ~ k "" Xnk , (34.18)
358 QUANTUM MECHANICS [Part 111
Thus, matrix elements depend harmonically upon time.
The energy difference A S n can be conveniently represented
by means of the Bohr frequency condition (see beginning of Sec. 23):
'**" m* kn . (34.20)
Therefore, the operator relationships (34.16) and (34.17), rewritten
for matrix elements, appear thus:
p n k = imuknXnk, (34.21)
* (34  22 >
The matrix form of the equations of quantum mechanics was
found by Heisenberg.
The probability for transition to a continuous spectrum. Let us now
investigate the expression for the transition probability (34.11),
rewriting it in matrix notation:
, ()  4 sin* <&=LpLL . . (34.23)
4 tl (GI <5 n)
In the examples dealing with radiation and ioiiizatiou that we
have mentioned, the final state of the system belonged to a continuous
energy spectrum. Indeed, the energy spectrum of an electromagnetic
field is continuous since the field can contain quanta of any frequency
o). In the ionization example, the spectrum of the electron emitted
from the atom was continuous because the motion of the electron
was infinite.
If the state n belongs to a continuous spectrum, it is more interesting
to determine the total probability of transition to any of the states
with energy S n , i.e., to find the integral of (34.23) with respect to
dS n . Now it is not advisable to state the final energy ff n since it
varies continuously; it is better simply to write . Then there are
dN (ff) states contained within the energy interval between $ and
ff\d. The example of dN () was given in Sec. 25, equation (25.25).
(The transition to a continuous spectrum in Sec. 25 is achieved
simply by means of an infinite increase in the box dimensions, which
are not contained in any physical result. The distance between neigh
bouring levels is then infinitely reduced.)
So let
dN(g)=z(g)d. (34.24)
The total probability of transition to the continuous spectrum is
r r A *(* *i> *
I / 48m 2*
= J w (/, , t) dN (t) = J w ~^
(34.25)
Sec. 34] THE QUANTUM THEORY OF RADIATION 350
For clarity in notation we shall put the indices <\, $ of JlfM in
brackets and not as subscripts, treating them as the arguments
of the function, which in fact they are with respect to
We shall denote the argument of the sine by the letter
2h " s *
Passing to the integration variable we obtain
(34.26)
The function ^ ^ has a principal maximum for = 0. Its next
maximum is already twenty times smaller. For this reason, in the
integral (34.26), the main part is played by the values of of the
order unity. But then the instant of time t can always be chosen
2/t y
so that " is considerably less than. tf\. Fn other words, it is per
missible, in the arguments of the functions ,^ (i) ($ v $ ) and z (<?'),
to replace g simply by <\ and to take out tho functions  ,#" CI) (c^,
^* ^j)! 2 and 2(^=^1) from under the integral sign. It is shown
thereby that if the time t is sufficiently long, the energies of the initial
and final states and S arc defined so accurately that they can be
considered simply equal to one another, in accordance with the Jaw
of conservation of energy in the transition. Naturally, tho law of
conservation of energy holds always, but, for sufficiently small
values of t, it is impossible to determine tho energy of the final state,
for the uncertainty relation (28.15) for the given case is of the form
($ $i) ^2 TcA. Hence, if t tends to infinity, the precise equality
ff=ff li is obtained.
Since the function ~ decreases rapidly with increasing 5, the
integration should be extended from oo to oo. Since the remaining
values have been taken out from under the integral sign, the integral
itself can be evaluated. It is
~dZ = n. (34.27)
00
From this
W^ AlL I jjf (i) (^ l) ^ == ^ t ) 12 2 (<> j . t t (34.28)
/i
Then the transition probability in unit time is
300 QUANTUM MECHANICS [Part III
We write the second argument S'=^ > 1 to emphasize that the state
fy (S ) coincides with <p (I x ) only with respect to energy. The formula
(34.29) has very many applications.
The matrix clement corresponding to the emission of a quantum.
With the aid of expression (34.29) it is possible to obtain rigorously an
expression for radiation intensity. This result is based on quantization
of the electromagnetic field performed in Sec. 27. We shall not give
the other, less rigorous, result based on the analogy between classical
equations and the equations for matrix elements.
In order to simplify subsequent computations, we shall, from the
start, take advantage of the law of conservation of energy for radiation.
In considering transitions of an atom from a state with energy ^ > 1
to a state with energy *f , we take only those quanta which satisfy
the energy conservation law in accordance with the Bohr frequency
condition S\ ^ =Aco. Further, we shall first of all consider quanta
with a definite direction of the wave vector k, and a definite polari
zation a. In addition, we assume that there Avere no quanta of this
type in the field initially, i.e., in the initial state N^ a = 0.
In this case the perturbation energy operator is the product of two
operators :
(vA). (34.30)
c
This expression is obtained from (15.32), if the term that is linear in A
is retained in the equation, and if we put cp 0.
The wave function of the atom and the field in a nonperturbed
state, i.e., with the interaction between them "switched off," is
expressed as the product of the wave functions of the atom and the
field. The wave function of the field is represented as the product of
the wave functions of separate oscillators with different k, o. All
these functions are orthogonal and normalized. Therefore, in calculat
ing the matrix element of the quantity Ag or of the coordinate of
the oscillator with given k, cr, we must take the wave functions
corresponding to the given oscillator. In accordance with the nor
malization condition, the integral over all the remaining coordinates
of the field gives unity.
In the vector potential we take the term relating to k and a :
(34  31)
where we have used equations (27.20) and (27.21) that express the
field amplitudes in terms of real variables.
We must calculate the matrix element describing the transition
between two states, with no quantum described by wave vector k
and polarization a in the first one, and with only one such quantum
See. 34] THE QUANTUM THEOUY OF RADIATION 3(51
in the second one. We shall write these states with subscripts and 1.
Then, from (34.21),
(Pg)oi = ico 10 (eg) 01  i<o k (#g) ol , (34.32)
since o> 10 is equal to the energy difference between the initial and
final states of the 1 field divided by h, i.e., just equal to the frequency
of the emitted quantum cot. Substituting this into the expression for
the matrix element (A) 01 we find that
' (34.33)
since the coefficient of e ikr becomes zero.
For simplicity we shall temporarily omit the indices k and o.
We must evaluate the integral
(34.34)
Here, the fieldoscillator wave functions are denoted by 9 am i 9i
in order not to confuse them with the atom wave functions.
We know the functions 9 and 9, from Sec. 26. From equations
(26.22) and (26.23) we have
1 u> Q 2 . i
< 34 ' 35)
The coefficients g Q and g l are found from the normalization condition
exercise 1, Sec. 26):
(e  "?)'  ,: f A Jrf *r *  : ]/' ; ,. = f ,: " ;
We note that the product Q<p Q is proportional to 9^ Hence, the
integral (34.34) would have vanished due to the orthogonality con
dition, if any other function, 9 2 , <p 3 , . . . , 9,,. had been substituted, in
place 9 X , into the integral. Therefore, only one quantum can be
emitted with a given frequency, direction, and polarization. The
same can also be shown for any arbitrary initial stage of the field.
Absorption of quanta also occurs singly. For Q ol we have
:  (34  36)
3012 QUANTUM MECHANICS [Part III
The dipolo approximation. It is now necessary to calculate the
matrix element given by two atomic states:
^ *o (v A 01 ) ^dV. (34.37)
Substituting A 01 from (34.33) and (34.36), we reduce this matrix ele
ment to the form
dV . (34.38)
The wave function of the discrete spectrum of an atom differs from
zero in the region near 10~ 8 cm, i.e., of the order of atomic dimensions.
The wavelength of visible light is about 0.5 x 10~ 4 cm, i.e., several
thousand times greater. Therefore, we can consider that the phase
of a wave changes very little in the region of the atom, and we remove
the exponential factor from the integration, taking it at some mean
point (for example, the nucleus).
This corresponds to a dipole approximation defined in Sec. 19,
the wavelength being considerably greater than the dimensions of
the radiating system. The other condition concerning the applicability
of the dipole approximation is that the electron velocity must be
considerably less than the velocity of light a thing that occurs
in atoms of small and medium atomic weight.
To the dipole approximation we have
V . (34.39)
From (34.18) the velocity matrix element is directly expressed
in terms of the coordinate matrix element
VIG = ^01 r io =  *co k r 10 , (34.40)
a> _ jp
because, according to the law of conservation of energy, co 10 =  v
is equal to the frequency of the radiated light.
The square of the modulus of the matrix element is
12 f . .
I e y l e k r lol 104.4IJ
We shall now take into account the fact that an emitted quantum
can have two different polarizations. If we are not specially interested
in the probability of quantum emission with a given polarization, then
the probability must be summed over the polarizations, i.e., over o.
To begin with, let us assume the vector k to be in the direction of the
zaxis. Then the unit vector e k can have two directions: along the
#axis and along the yaxis. Accordingly,
Sec. 34J
THJS QUANTUM THEORY OF RADIATION
2 =1*. J24U/.J2
Let us find the average of this expression over all possible directions
of quantum emission. It is then obvious that
(34.43)
By performing this averaging after summation with respect to cr,
we obtain
In order to find the probability of emission of a quantum in unit
t)
time, we must multiply (34.44) by Vz (). z (S 1 ) is found from
equation (25.24), where we must put k~  :
c
c ) _. 2 /jy* 4rt\
Finally, from equation (34.29), we find the expression for the
probability to a dipole approximation:
dir
~dt~
4 co 3
y^c a ~
(34.46)
We can write the product er 10 as d 10 , i.e., as the dipole moment
matrix element.
The intensity of radiation is equal to the radiation probability in
unit time multiplied by the energy of the quantum:
This equation greatly resembles the classical formula (19.28). How
ever, we have the square of the modulus co 4  d 10  2 in place of the
square of the second derivative of dipole moment d 2 . The correspond
ence between classical and quantum theory displayed here may be
demonstrated by means of the matrix equations (34.16) and (34.17)
too. Directly applied to electrodynamics, it leads to equation (34.47).
We have given a more rigorous deduction, based on the quantization
of electrodynamical equations, in order to illustrate the generality of
the methods of quantum theory.
Compared with the classical formula (19.28), the quantum expression
contains an extra factor of two (4/3 instead of 2/3). This is explained
364 QUANTUM MECHANICS [ L'cirt 111
in the following way. We represent a classical dipole moment, varying
harmonically, in the following manner:
(1 = d x e"' )f + d} e"''<' , d = co 2 (d x f"' M ' + d} e~ /wf )  (34.48)
The terms d x e /w/ and dj e~' wf depend upon time like the matrix
elements d 10 and d 01 . Let us form the time average of (d) 2 . The terms
involving e 2 '"' and e~ 2iwr drop out in averaging, and there remains
But it is the quantum formula which corresponds to the mean
radiation intensity I Aco '. 1 , so that the factor of two is due to the
time averaging of the square of the dipole moment, given in the
form (34.48).
The expression (34.47) confirms what was said in Sec. 22 about an
atom being stable in quantum theory: radiation is always associated
with a transition of the atom from one state to another. But no
atomic state exists for which the energy is less than the energy of the
ground state, that is why the atom can exist in the ground state for
an indefinitely, long time.
The selection rules lor the magnetic quantum number. It follows
from equation (34.41) that, if r 10 = 0, the intensity is equal to zero,
at any rate to a dipole approximation. We shall now find the con
ditions for which r 10 differs from zero. First of all we notice that if
the vector defining the polarization direction is equal to e then, to
a dipole approximation, radiation of such a quantum is possible
provided the matrix element for the projection of the electron radius
vector along the direction of quantum polarization differs from zero.
Let us assume that the quantum polarisation is along the zaxis.
Let the magnetic quantum number of the electron be equal to k
before the transition, and k f after the transition. Then the dependence
of the wave function upon azimuth ^ is given by the equations
<I>i Mr, )**, *;=/;(r,&)e^,
because e ik * and e tk 'v are eigeiifuuctions of the angular momentum
z projection. Hence
27t
z 10 = J/J (r, ft) r cos ft  f 1 (rb) r 2 sin 9 dr dftjV* (k ~ fc '> d 9 ;
o
a*'),/  ei * (k ' fc/) I"" I for & ^ fc ''
d(?  ~i II  k'} \  \ 2 TT for fc = k' .
0
2 ~r cos ft in polar coordinates and, therefore, does not depend
upon 9. For this reason, the matrix element z 10 differs from zero
only when k' = k.
Sec'. 34] THE QUANTUM THEORY OF RADIATION 365
Instead of considering plane polarized radiation along x or y, let
us take circularly polarized radiation in the #yplane. In such radiation
there is a constant phase shift between the xtii and the yth components
equal to ^ (see sec. 17, Fig. 25). Consequently, we must determine the
matrix elements of the quantities
. TC
(x\e * 2 y) lQ = (x { iy) 1Q = (r sin&e <<% .
Substituting the expressions for the wave functions involving 9
explicitly, we find
27T
V<fcfc':t D<Pd 9 ^'i'^r  0, if jfc' ^ fc 1 . (34.49)
o ~^
Hence, radiation which is circularly polarized in the #?/plane can
only be emitted if the magnetic quantum number changes by 1.
The rules that determine what change of quantum number governs
the emission of a given radiation are called selection rules.
The selection rules for dipole radiation with respect to the magnetic
quantum number forbid the changing of k by more than unity.
The selection rules for the azimuthal quantum number and parity.
The magnetic quantum number is the angularmomentum projection.
Since the angularmomentum projection does not change by more
than unity, the angular momentum itself (i.e., the azimuthal quantum
number) cannot change by more than unity.
But I for a separate electron cannot remain unchanged, because
then the functions d^ and A must have the same parity. Here, the
product fy*Q z fyi will turn out to be an odd function while its integral,
i.e., the matrix element j ^zfy^dV, will become identically zero.
In exactly the same way f Jj;J x^ d V and \ tybyfyidV will also become
zero. This is why, for a dipole transition of one electron in the atom,
the azimuthal quantum number changes by 1.
Angular momentum and parity of a light quantum. As was indicated
in Sec. 13, an electromagnetic field possesses angular momentum. If
from equation (13.28) we determine the angular momentum of a
quantum emitted during dipole radiation, it comes out equal to unity.
And the state of the quantum is odd because it is determined by the
parity of the dipolemoment vector components d, which, obviously,
change sign for the interchange x> x, y> y, z> z. Hence, the
selection rules for the azimuthal quantum number and parity of the
state of the atom must be interpreted as the conservation laws of
total angular momentum and total parity of the atom + quantum
system in radiation. Clearly, if the angular momentum of a quantum
306 QUANTUM MECHANICS [Part III
is equal to unity, the angular momentum of the atom cannot change
by greater than unity during radiation.
The selection rules for spin and total angular momentum. If the spin
is in no way related to the orbital motion, the spin functions for the
initial and final states must be the same, otherwise the transition
dipole moment is equal to zero due to the orthogonality of spin func
tions that correspond to different spin eigenvalues.
This selection rule is approximate in character and is valid for light
atoms. Taking into account the spinorbital interaction, we must
consider the selection rules for the total angular momentum j =Ma
[see (22.15)]. Since the angular momentum of a dipole quantum is
equal to unity, we obtain the condition for j' j: j'=j or j'=jl.
Here, the parity of the state must change. However, since the parity
is not directly related to j but only to Z, the transition j' =j is also
possible. But the transition from j = to j' is forbidden, because
in this transition the quantum cannot acquire the angular momentum.
It is necessary to note that the angular momenta for quanta of higher
multipole order than dipole can only be greater than unity, so that the
transition from j = to j' = is forbidden for all approximations, and
not only to the dipole approximation.
The selection rules for manyelectron atoms. By considering a light
quantum as a particle with unity angular momentum, it is easy to
obtain the selection rules also for cases when the states of more than
one electron change. Neglecting the spinorbital interaction, the
selection rules are the following: 8' =S, L' L or L'==JD1 and the
parity is reversed. The transition L' L is possible here, because, in
a manyelectron system, parity is not related to total angular momen
tum.
Magnetic dipole radiation. A system of charges may radiate as a
magnetic dipole as well as an electric dipole. Magnetic dipole radiation
is usually related to a change of spin projection k a . Since the spin of
an electron is onehalf, the angular momentum of an atom changes
by unity for a "flip" of the spin of an electron and for an unchanged
orbital angular momentum. The moment of a magnetic dipole quantum
is equal to unity just like the moment of an electric dipole quantum.
But the parities of the electric and magnetic quanta are reversed.
Indeed, the components of electric dipole moment change signs in an
inversion of the coordinate system (31.35), while the magneticmoment
components do not change signs because the magnetic moment, like
the angular momentum, is a pseudovector (see Sec. 16).
As was pointed out in Sec. 19, the intensity of magnetic dipole ra
diation is less than the intensity of electric dipole radiation, their
ratio being l~\ , where v is the charge velocity and c is the velocity
of light. This ratio is about 10* 5 for light elements.
Sec. 34] THE QUANTUM THEORY OF RADIATION 367
Quadrupole radiation. In Sec. 19 it was shown that radiation is
possible due to the change of quadrupole moment for the system.
Here, electric quadrupole quanta occurring in an even state are radiat
ed because the electric quadrupole moment is an even coordinate
function. The angular momentum of a quadrupole quantum is equal
to two.
Quadrupole radiation can occur when dipole radiation is forbidden
by the selection rules. From Sec. 19, quadrupole radiation is obtained
when taking into account the retardation inside the system. The order
of magnitude of this retardation is determined by the ratio of the di
mensions of the system to the wavelength of the emitted light. There
fore, the probability of quadrupole radiation is less than the proba
bility of dipole radiation in the ratio II , where r is the size of the
system.
X~0.5x 10~ 4 cm for visible light while the atomic dimensions are
r~0.5x 10~ 8 cm. Therefore, in order of magnitude, the probability
of a quadrupole transition is 1C 8 times less than the probability of a
dipole transition.
Metastable atoms. If an atom can go from an excited state to the
ground state only by means of a transition which is forbidden in dipole
radiation, it remains excited considerably longer than for a dipole
transition. For a strong forbiddence it may remain excited for a very
long time (even on the ordinary scale, and not the atomic scale).
Such an atom is termed metastable. Usually, in gases which are not
highly rarefied a metastable atom gives up its excitation energy to
other atoms in collisions and not by means of radiation. Radiation will
then not be observed. But in highly rarefied matter, for example in
the solar corona or in a gaseous nebula, the spectral lines due to the
deexcitation of metastable atoms are very bright. For example, in
the spectra of nebulae, there occurs an intense magnetic dipole line
of doublyionized oxygen atoms.
Nuclear isomerism. Transitions with very large A; (up to 5) are
observed in nuclei. For small excitation energies, of the order of several
tens of kilovolts, metastable nuclei have very large deexcitation
times days or months. Such nuclei are called isomers with respect
to the basic unexcited state of the nucleus. The phenomenon of nuclear
isomerism in artificially radioactive nuclei was first discovered by
I. V. Kurchatov and L. I. Rusinov (in Br 80 ).
The totally forbidden transition. The transition from j = to j' = 0,
with an energy of 1,414 kev is observed in the RaC nucleus. Since the
radiation in this case is completely forbidden, the nucleus simply
ejects an electron from the inner atomic shell by means of an electro
static interaction; this may be explained as follows.
If an internal nuclear rearrangement occurs, the charge distribution
inside it somehow changes. For a 0>0 transition, one spherically
308 QUANTUM MECHANICS [Part III
symmetrical charge distribution is rearranged into another, which is
also symmetrical, but with a different radial dependence. Therefore,
in accordance with Gauss' theorem, only the electric field inside the
nucleus is changed. The field outside the nucleus cannot change; for
instance, it cannot radiate quanta. The wave functions of the ^states
of the electrons differ from zero in the nucleus. It follows that a change
of field inside the nucleus is capable of influencing an electron and im
parting to it an energy sufficient for ejection from the atom. In accord
ance with the law of conservation of energy, the electron, upon
ejection from the atom, will have an energy equal to the energy differ
ence of both spherically symmetrical states of the nucleus minus
the binding energy in the atom.
It may be stated generally that the ejection of electrons from an
atom shortens the lifetime of metastable isometric nuclei, since it
makes transitions more possible.
Sec. 35. The Atom in a Constant External Field
A classical analogue. In considering the behaviour of a system of
charges situated in an external magnetic field, it is very convenient
to proceed from the idea of the Larmor precession of magnetic moment
around the field. The only component of the angular momentum con
served in such precession is that directed along the field, both trans
versal components averaged over the precessional motion being
zero.
The situation in quantum mechanics is analogous, with the differ
ence that the projections perpendicular to the field do not exist as
physical quantities. In this way a simple correspondence is established
between the integrals of classical and quantum mechanics. The angu
larmomentum projection on the magnetic field is such a corresponding
quantity; it can be called a quantum integral of motion.
* An external magnetic field superimposed on an atom perturbs its
state in a definite way. The Hamiltonian operator for such an atom
may be divided into the operator Jf^ for the unperturbed atom and
the perturbation operator J6 (1) due to the magnetic field.
Addition of magnetic moments. Let us first of all write down the
operator JT (1) explicitly. It was shown in Sec. 32 that spin motion does
not produce the same magnetic motion as orbital motion, namely, the
magnetic moment for orbital motion is
and the spin magnetic moment
(35.2)
\ )
Sec. 35] THK ATOM IN A CONSTANT EXTEHNAL tflKLl) 309
Therefore the total magnetic moment is
(35.3)
Hence, the magnetic moments are not combined according to the
same law as mechanical moments:
JL + 8. (35.4)
Comparing (35.3) and (35.4), we see that the magnetic moment of an
atom is not proportional to its mechanical moment.
In accordance with (15.35), the perturbation energy caused by the
magnetic field is equal to (the moments are expressed in h units)
= (L f 28) = tioH(J + 8) . (35.5)
Here X = j^ is the Bohr magneton. The plus sign resulted
because the charge of the electron is e. We note that the magnetic
energy in expression (15.35) was defined as a correction to the Hamil
tonian, i.e., to the energy expressed in terms of momenta. Therefore,
in quantum theory it is directly interpreted in terms of operators.
The vector model of the atom. To a first approximation, the energy
correction is equal to the mean value of the perturbing energy taken
over unperturbed motion [see (33.31)]. Therefore, we first find the
unperturbed state of the atom without the superimposition of a magnetic
field. Let us suppose that a normal coupling exists in the atom
between the total spin and the total orbital angular momentum
(see Sec. 33), i.e., all the orbital angular momenta of the electrons are
combined in one resultant orbital angular momentum L, and all the
spin angular momenta are combined in one resultant spin angular
momentum 8. Examples of such orbital and spin angularmomentum
composition were given in Sec. 33 (in the text and in the exercises).
For example, in combining the angular momenta of two r&pelectrons,
the following states are obtained: 1 Z), 3 P, and 1 S. All these states are
formed in accordance with the Pauli principle, and possess spatial
wave functions of different forms. For this reason, in all three states,
the energy for purely electrostatic electron interactions differs by
magnitudes of the order of an atomic unit, i.e., by several electron
volts.
Let us choose the ground state of these states. In accordance with
Hund's first rule, this is the 3 P state. We have not written the sub
script J here because it can have three values: J = 2, J= 1, and / = 0.
Accordingly, we have written 3 on the upper left. The states which
differ only in J, for identical L and S, are considerably closer to each
other than the three states with differing 8 or L listed above.
Let us estimate the order of magnitude for multiplet level splitting,
i.e., the spacing of levels with different J. A magnetic field of moment
370 QUANTUM MECHANICS [Part III
(x is of the order ~ , so that the interaction energy of two moments is
2 ^*
3  . To evaluate the order of magnitude we put one Bohr magneton
in place of JJL, i.e., 10~ 20 , and r^0.5 x 10~ 8 . This results in an inter
action energy of the order 1Q 15 , i.e., thousandths of an electronvolt
(in practice, greater multiplet splitting is observed due to larger [i
and smaller effective radius values). In any case, the levels 3 P 2 , 3 P 1?
and 3 P , which are comparatively different from the other two levels
1 D and 1 S 9 occur close to one another. The 3 P level is split into three
finestructure levels which, in the given case, corresponds to the super
script 3. If L<S, the number of components of the multiplet splitting
is determined by L.
Each of the levels with a given J corresponds to a definite configu
ration of the vectors L and S. In classical theory, we would say that
L and S are parallel in the state with J = 2, antiparallel for J = 0,
and perpendicular for J1. Of course, the latter one is not at all
meaningful in quantum theory because only one angularmomentum
projection exists. The projection of S on L is equal to zero for J = 1,
and the other projections do not exist.
At the beginning of this section we indicated that, in the classical
analogy, those components which are not conserved are, in some way,
averaged over the Larmor precession of angular momenta and yield
zero. In this case, we are not concerned with precession in an external
magnetic field, but with that in the internal field of the magnetic
moments themselves. Since J is an exact integral of motion we can, in
a visual demonstration, consider that the direction of J is fixed in space,
while the triangle consisting of the vectors L, S, and J processes about
J in space. In the cases for which J = 2 and J the triangle degen
erates to a straight line. Thus, to each of the multiplet levels there
corresponds a definite vector model given by L,
S, and J. We note that this refers to normal
coupling.
An external magnetic field H causes the
vectors L, S to precess about its direction. It
is most simple here to consider the two opposing
limiting cases. We shall examine them.
A weak external field. Let the external field
be weak compared with the effective internal
field that the multiplet level splitting is due to.
Fig. 44 Since the Larmor precession frequency is propor
tional to the magnetic field, the triangle LS J in
this case rotates about the side J considerably faster than the precession
about H. During the time of one rotation about H, the triangle can
rotate very many times about J. Therefore, the coupling of the vectors
L, 8, and J in the triangle is not disrupted, as it were, due to the
internal magnetic forces forming the triangle being large compared
Sec. 35] THE ATOM IN A CONSTANT EXTERNAL FIELD 371
with the external magnetic force. We have shown this idea in
Fig. 44.
Let us now find the correction due to the magnetic field. In calculat
ing the mean value of the perturbation energy from the unperturbed
motion, it is convenient to make use of the Larmor precession model.
In this case, two forms of precession must be considered: the triangle
LSJ about i and the precession of J about the magnetic field.
Equation (35.5) involves the vectors J and S. It is very simple to
average J: we must take its projection on the magnetic field J z . We
shall consider that the 2axis coincides with the direction of H. The
projection of 8 upon H is not meaningful because the vector S together
with the triangle LSJ rotates considerably more rapidly about J
than about H. The component S, perpendicular to J, is averaged by
the precession in motion unperturbed by the external field. There
remains the projection parallel to J and equal to
(35.6)
/O J\
Obviously, the projection of this vector upon H is equal to J x ' .
Thus, the mean value of JdfM is proportional to J x and is equal to
. (35.7)
It is now necessary to give a quantum meaning to the product
(SJ). From the definition of J (35.4) we have
L = JS. (35.8)
Squaring this equation, we get
(35.9)
Expressing the square of the angular momentum in h units, in accord
ance with (30.29), we have
Let us make similar substitutions for J 2 and S 2 . Therefore
(SJ)
~~
Substituting (35.10) in (35.7) we obtain finally
. (35.11)
Thus, the finestructure level with a given J is split into as many
levels as there are different projections of J on the magnetic field,
372 (JUANTTM MECHANIC'S [l^U't III
i.e., 2J + 1 levels. For given L and 8, the following definite factor
corresponds to each value of J:
,,_! t
g , 1  h
It is called the Lande factor.
For example, for L~/S y J = l, we obtain
Analogously, for /=2, 8~L = l
7== 1 + i :?. 2. = : <
^ X 1 2 6 2 '
Tho level with J^O does not split.
Splitting in a strong field. Tho representation of splitting set out
here corresponds to reality only as long as the magnetic field is so
weak that the spacing between the 2 J + l levels of [L Q yHJ z in the
magnetic field is small compared with that between the unsplit multi
plet levels themselves with differing J. When the splitting in the mag
netic field is comparable with that of the multiplet itself, or is somewhat
greater, the pattern becomes more complicated, but in a strong field it
once again becomes very simple.
Therefore we shall consider the opposite extreme case, when the
external field is strong compared with the internal field, so that
the coupling between the vectors L, S, 3 in the triangle is disrupted.
The necessity for this disruption in a sufficiently strong magnetic
field can bo explained by the fact that S processes twice as fast as L.
Then, from the classical analogy, each of the vectors S and L processes
independently about the magnetic field, so that the correction to
the energy is given by a different expression from (35.11):
?U) = jff(V = ^eH(L z + 2S Z ) . (35.13)
Here L z is the projection of the orbital angular momentum upon
the zaxis, and S z is the totalspin projection of the atom upon the
same axis (in h units). Naturally, the total values of L and S are not
changed by the magnetic field, though the distribution of levels
in a strong magnetic field is not related to the multiplet structure,
as was the case in a weak field, but only with the possible projections
of L and S on the magnetic field. The vectors L and 8 precess about
the field far more rapidly than they precess about 3 without the
field. This is why the coupling in the triangle is disrupted.
The projections S z and L z are changed by unity, therefore all
the levels #W in expression (35.13) are equidistant. Of course, certain
values of gM may be repeated several times if the sum L Z + 2S Z
assumes the same value in several ways. For example, if L = 1, #== 1,
tSec. 35] THE ATOM IN A CONSTANT KXTEHNAL F1KLD 373
then we get the following range of values of the sum : 1+2 = 3,
+ 22, 1+0 = 1, 1+2 = 1,0 + = 0, 1+0= 1,1 2= 1,
2 2, 1 2= 3; there are in all seven equidistant
values, and 1 and 1 are obtained in two ways (i.e., each of them
from the confluence of two levels), so that there are nine states in
all. We note that in a weak field the same multiplet split thus : J = 2
into 5 levels, J = l into 3 levels and J = did not split. As was to
be expected, the total number of different states in the strong and
weak fields is the same.
The radiation spectrum for level splitting in a strong field. Let us
now see what spectral lines appear when light is emitted from an
atom situated in a magnetic field. To begin with, let us consider a
strong field, because the pattern of the splitting of spectral lines is
simpler in this case than in that of a weak field. Both levels, upper
and lower, resulting from two multiple ts are split into a certain
number of equidistant levels in accordance with formula (35.13).
Let the radiation be observed in a direction perpendicular to the
magnetic field. The radiation polarization vector is perpendicular
to the direction of propagation, i.e., it is either directed along the
magnetic field or in a third perpendicular direction, say along the
a:axis (the magnetic field is along the zaxis). The selection rules
for radiation polarized along z and along x are different. For polari
zation along the zaxis, the orbital magnetic quantum number must
be conserved. S z is also conserved for all polarization in. which the
spinorbital interaction is neglected. Therefore, all Hues polarized
along the zaxis, i.e., along the magnetic field, have the same frequency,
which corresponds to the energy difference of the two initial levels
<^i <^o prior to splitting in the magnetic field: the correction (35.13)
is cancelled in calculating the difference <?W ^j, n . A wave pol
arized along the #axis can be represented as the sum of two waves
circularly polarized with opposite directions of polarization. Tho
selection rule for these lines is that k can change only by _L 1. Con
sequently, the radiation polarized along the #axis has a frequency
that differs from the initial frequency by db  . In observing
the spectral lines emitted perpendicularly to a strong magnetic field,
the original line is thus split into three lines separated by an interval
which is equal to the Larmor frequency for the given field.
If we drill a hole in the shoe of an electromagnet it is possible to
observe radiation propagated along the magnetic field. It is circularly
polarized in the #yplane. The selection rules for right and lefthand
circular polarization correspond to a change of k by 1, so that
there will be observed two lines spaced from the centre by ~ .
< fY\i C
Thus, when the field is switched on, the original line will split into
two Ijnes separated by an interval equal to twice the Larmor frequency.
374 QUANTUM MECHANICS [Part III
Exactly the same picture is found in the classical oscillatory motion
of a charge situated in a magnetic field. This problem was considered
in exercise 6, Sec. 21.
The effect of the splitting of spectral lines in a magnetic field
was discovered by Zeeman before the quantum theory of the atom
appeared. Therefore, the then accepted theoretical explanation of
the Zeeman effect corresponded to the classical problem, where it
was considered that the charge performs an oscillatory motion.
However, in observing spectra, this classical picture applies only
in strong magnetic fields such that the splitting of lines obtained
is considerably greater than the spacing between multiplet levels.
Under these conditions the Zeeman effect is termed normal, because
outwardly it corresponds to the theoretical ideas of the time at which
it was discovered. It may be noticed that a field which is strong
for one multiplet can still be weak for another.
Spectralline splitting in a weak magnetic field. The Zeeman effect
in a weak magnetic field is termed anomalous. A spectral pattern
is obtained which is entirely different from the classical. First of all,
the number of splitting components can differ from the normal.
The distances between them are also quite different.
As an example let us consider the anomalous Zeeman effect in
the socalled Dline doublet of sodium. This line is double without
an external magnetic field. It corresponds to the two transitions
2 Pi> 2 $t and 2 Pa> 2 $i. The 2 P level has an orbital angular momen
2" 2 1 ~2 2~
turn 1 and spin ^ . Therefore, the resultant value of the total angular
13 11
momentum J can be 1 + ^ =   and 1  ~  5  . This is where we
2t 2i 2i Zt
get the fine doublet structure of the 2 P level in the absence of an
external field. The 2 $ level cannot split without a field because it
has an orbital angular momentum of zero. The double Z)liiie in the
sodium spectrum arises in the transition from the doublet level
to the single. According to our rough estimate of the finestructure
splitting, the difference in frequency between its components amounts
to about one thousandth of the mean frequency of the doublet.
The 2 Px level is lower than the 2 P^ level.
2 2
Let us now calculate the Lande factor for three levels.
JL V* Vi + V* V.  i il _ _*
2 ~7, /,* ~ 3
2)
Sec. 36 J THE ATOM IN A CONSTANT EXTERNAL FIELD 375
3) &,,: J = 1 lt, i = 0, 5 = Vi.
rt14. J VtVi + VlVl _0
(71 + T  i^  2.
In accordance with equation (35.11), we have an expression for
the energy of the 2 P_a_ state in a magnetic field. For conciseness we
shall denote the quantity [L Q H by the single letter p. Then
But J z takes on four values: 3/2, 1/2, 1/2, 3/2. Hence, in a field,
the *P level splits into four levels, whose energy differences from
the central, unperturbed, state are
respectively.
We obtain two energy values for the 2 Pj^ finestructure level:
2
V 2 ) = P
And, finally, for the lower 2 S level, we get
2
Let us now find the spectral pattern. We start with the 2 P_^ > 2 S
transitions. The oscillations polarized along the field obey the se
lection rule AJ* 0. Hence, their frequencies are shifted relative
to the central position by
and by
/, , V 2 )  P ~ P = 
Unlike the normal Zeeman effect, a double line has been obtained
also for radiation polarized along the magnetic field.
For perpendicular polarizations we have
/.) = JPP 4e
 * 
J P + P = 4
376
QUANTUM MECHANICS
[Part III
Let us take the transition 2 Pjr* 2 &* If the oscillation is polarized
2 2
along the field we once again have, of course, two lines, though with
other spacing:
V.)
,,, , _ i/ a) =  3 p + p = y p ,
> Vi) =  P  P   P 
We have, for both circular polarizations:
These are the results for righthanded polarization. The corresponding
rj
splitting for lefthanded polarization is p and ^ p.
Thus, one component of the ZMine is split into six Zeemaii compo
nents, and the other into four.
In the given case, the Zeeman effect remains anomalous as long
as p is negligibly small compared with one thousandth of a volt,
or the magnetic field is very much
* A loss than 5,000 CGSE units.
A diagram of the splitting is shown
in Fig. 45.
The atom in an electric field
(Stark effect). The multiplet levels
*/fe ,_,,_ . % f r a certain total angular momen
~ J /z turn J split in an electric field, too.
tt>U) We shall consider first of all the
a/o!) ' ' ' J case of a weak field, when the level
*P % *~*Si shift caused by the field is small
compared with natural multiplet
splitting.
First of all, we must bear in
t mind that the angularmomentum
projection on the electric field is
determined only within the accuracy
Fig. 45 of the sign, because the angular
momentum is a pseudovector while
the electric field is a real vector. In reversing all the coordinate signs,
the angularmomentum components change sign while the electric
field components do not change sign. But since the choice of right
handed or lefthanded coordinate system is arbitrary, the projections
of the angular momenta on the electric field are physically deter
mined only to the accuracy of the sign. If J is an integer, the number
~&
Sec. 35J THE ATOM IN A CONSTANT JSXTKRNAL FIELD 377
of its projections which differ in absolute value is equal to J+l
(0, 1, ...,/), while if J is a halfintegral number then the total number
of projections is : J f y ly , y , . . . , Jl . For example, if J = y, there
is only one nonnegative projection. Therefore, the state with angular
momentum ^ is not split by an electric field, at any rate as long
as the coupling between L and S is not disrupted. For comparison
we note that the magnetic field splits the state witli J~ 9 into two
states because the magnetic field, like the angular momentum, is
a pseudovector.
In a stronger electric field the coupling between L and 8 is disrupted.
In this case the scheme of splitting is the following. The vector L
is integral. It has L + l projections on the electric field. We must
project 8 onto its projection. But since L and S are both pseudovectors,
the number of projections of S upon L is already equal to 2$ + l.
The only exception is when the projection of L upon the field is equal
to zero. This level splits into $ + 1 or $+ levels according to 8.
The squarelaw Stark effect. The amount of splitting is determined
by the relative shift of neighbouring levels. As was shown in Sec. 33
(33. 31 a), the shift of an energy level is equal to the average of the
perturbation energy for unperturbed motion. Proceeding from (14.28),
we have the following expression for the perturbation energy in
a homogeneous electric field
(35.14)
But it is easy to see that the average of this quantity is equal to
zero. Indeed, the wave function of an atomic state with given /
is always odd or even (with the exception of hydrogen, see below).
Therefore, the product t^y fyj must be even. From (30.24), the average
of ^f^ is equal to
(35.15)
But the integrand is an odd function, so that its integral is identically
equal to zero.
Level splitting is obtained only to a second approximation, if
into (35.15) we substitute wave functions which have already been
perturbed by the external field. This splitting is governed by a square
law field dependence.
The linear Stark effect. In a hydrogen atom the electron energy
depends only upon the principal quantum number n and does not
depend upon L Therefore, the state with <?=<f n is represented as
a superposition of states with I varying from to n 1. But the wave
function is even for even /, and odd for odd L Hence, the function
378 QUANTUM MECHANICS [Part III
with $ = & n does not have a definite parity, so that the integral
(35.15) does not become zero. Therefore, in the hydrogen atom we
observe line splitting which depends linearly upon the electric field. *
Highly excited atomic states always more or less resemble hydrogen
atom states, because the nucleus and the atomic residue act upon
an electron, which has receded far from the nucleus, in a way similar
to a point charge. The energies of these states depend upon I in ac
cordance with the expression (31.46). These states give a linear
Stark effect if the perturbation produced by the field shifts the levels
more strongly than they are split in I.
lonization of the atom by a constant field. A constant electric field
not only shifts the energy levels of an atom, but also qualitatively
changes its whole state.
Let us write down the potential energy of an electron in an atom
situated in an external electric field E which is directed along the
zaxis :
U = U Q (r)+eEz. (35.16)
For a sufficiently large and negative z the potential energy far away
from the atom is less than in the atom. The potential well in the atom
is separated from the region of large negative z (where the potential
energy can be still less) by a potential barrier. But there is always
the probability of a spontaneous electron transition through the
potential barrier into the free state. Transitions of this type were
considered in Sec. 28 as applied to alpha disintegration.
Any state of an atom put in a constant electric field may be ionized,
but, naturally, if the field is weak the probability of ioiiization be
comes vanishingly small. In a strong field the potential barrier be
comes transparent, especially for highly excited atomic states. If
the time for the spontaneous ejection of an electron in such a state
turns out to be less than the radiation time, the corresponding line
in the spectrum disappears.
Thus, a weak perturbation inside an atom (the atomic unit of field
intensity E = ^4 5.13 10 9 v/cm, so that the external field is
always small compared with the atomic field) essentially affects
the state since the conditions at infinity change. But if the broadening
of the atomic levels is still small compared with the distance between
them, they can be regarded, as before, as discrete.
Exercise
Construct a diagram for the splitting of the multiplet 3 JP> 3 and the
transitions in a strong and weak magnetic field.
* The relativistic expression (38.28) for the energy of a hydrogen atom
involves n and /. The orbital angular momentum I = / */ 2 for a given /,
so that a state with given n and / (in the same way as to a nonrelativistic
approximation) does not have definite parity and yields a linear Stark effect.
Sec. 36] QUANTUM THEORY OF DISPERSION 379
Sec. 36. Quantum Theory of Dispersion
The classical theory of dispersion, a brief outline of which was
given in exercise 19, Sec. 16, proceeds from the concept of a charge
elastically bound in an atom. The forced oscillations of these charges
under the action of a sinusoidally varying field lead to an electrical
polarization of the medium proportional to the field. Whence the
dielectric constant can be easily calculated as a function of the fre
quency.
The classical theory of dispersion is in good agreement with ex
periment. Yet, at the present time it is well known that the charges
in atoms are by no means bound by elastic forces. For this reason,
the success of classical dispersion theory may appear incomprehensible.
Even though the charges are not bound by elastic forces, there exist
quantities, relating to the motion of the charges, which vary har
monically with time: these are the coordinate matrix elements
[see (34.18)]. Similar harmonic oscillations occur, as is well known,
in the classical mechanics of elastically bound particles. The dipole
moment of an atom, induced by an external alternating field, is
expressed in terms of the dipolemoment matrix elements directly
related to the coordinate matrix elements. In the present section,
a quantum theory of dispersion will be formulated which will lead
to the same expression for dielectric constant as classical theory;
it will also indicate which quantities should correspond to each
other in both theories.
The wave equation for an atom in a given field of radiation. In
order to calculate the dipole moment induced by a field, we must
first of all determine the wave function of the atom in the external
field. In contrast to the previous section, where the behaviour of
an atom in a constant external field was studied, we shall here con
sider the interaction of an atom with an alternating external field
which varies according to the law
E:=E cosco*. (36.1)
It turns out to be more convenient here to write down the field,
straightway in real form instead of taking the real part of the final
result in order to have a real Hamiltonian.
The wavelength of a light ray is rightly considered large compared
with atomic dimensions (this was confirmed by the estimate in Sec. 34),
so that the field E may be considered homogeneous: its phase is
constant over the whole atom.
We determined the energy for a system of charges in an external
homogeneous field in Sec. 14 [see (14.28) and (35.14)]. The correction
to the Hamiltonian due to a homogeneous electric field looks like
(dE). (36.2)
380 QUANTUM MECHANICS  Part III
If we call the Hamiltonian of an unperturbed system fW, then
Schrodinger's equation will be of the form
Separating the wave function into an unperturbed part ^<) and
a perturbation ^ <1) > and regarding the perturbation as relatively
small, we obtain an equation which we have already used in Sec. 34:
_  
% (j t
Expansion in eigenfunctions. We seek the unknown function i]/ l >
in the form of a wave function expansion with timedependent co
efficients :
j,a>=2>^ 0) . (36.5)
n
We obtained an equation in (34.7), Sec. 34, for the expansion
coefficients
**= f W&^dV. (36.6)
The righthand side of this equation depends upon time in a different
way from that in equation (34.7), because the perturbation operator
jfrM involves time explicitly [see (36.1)]. Let us consider that the
unperturbed state of the atom is its ground state, which we shall
write with a subscript 0, i.e., ^ <>. Then there will simply be the
matrix element d ofl on the righthand side of (36.6) multiplied by
E cos w. The time dependence for the matrix element was found
in Sec. 34. Using the notation (34.20) we can write
dow^e'^iiod'oi,. (36.7)
The representation of a matrix element together with its time de
pendence is termed the Heisenberg representation, and that without
the dependence, in the form d' , is the Schrodinger representation.
Substituting (36.7) in (36.6) we obtain
 TlfiT = ~ Y (e ' (M + Wfld>/ + C ' (WI " w) (Eod'on) . (36.8)
In order to integrate this equation we must impose a certain initial
condition upon c. It is natural to suppose that the external field
acts for a sufficiently long time so that ail the transition processes
related to turning on the field do not affect the states. We can assume,
for example, that the external field depends upon time according
to the law:
Sec. ,'J()J QUANTUM THEORY OF D [Sl'KItSlOM 381
cos6> for <0,
,
EEocoseo* for *^0, ( '
i.e., the amplitude gradually rises with time to the value E . This
law for the change of the field must be substituted into (36.8), inte
gration performed from oo to any t, and a must tend to zero.
After this, at each instant of time (t < or > 0) there will be a single
dependence of c n upon t:
i / P l (a) *jO ~ ) '
Cn = ~  h
2h \ w /iO~ w
Induced dipole moment. The mean value of the dipole moment
is calculated according to the general formula (30.24) for mean values:
= J
(36.11)
The quadratic term in <J/ l > must, of course, be discarded, since
the calculations are performed to the accuracy of terms proportional
to E in the first degree. In addition, the term j^(o)*d<J/ 0) ^F does
not depend at all upon E and, therefore, is irrelevant to the problem
of polarization produced b> an external field. Also, this term is usually
equal to zero, as indicated in the previous section in connection with
the expression (35.15). Hence, the mean dipole moment responsible
for dispersion is
d = J (4/<>)* d^ 1 ) + ^( 1 >*d^())dF. (36.12)
We shall substitute here the expansion (36.5) and integrate the
series term by term:
a =
The integrals involved here are once again dipole moment matrix
elements. Substituting their expressions from (36.7), we write the
mean dipole moment as
d~  JW "to*' d', l0 + c*e" 1>0 "' d' ,,) . (36.14)
n
With the aid of expression (36.10), we finally obtain for c n :
(36.15)
w/,0 *} eo
Here we could already have written d in place of d^ , because the
time factors of d,, and d 0/J cancel.
382 QUANTUM MECHANICS [Part III
Polarization. In order to calculate the polarization of atoms by
a lightwave field, it is sufficient to know only the dipole moment
projection on the field. If, for example, the electric field of an incident
wave is directed along the #axis, then the expression (36.15) involves
only angularmomentum transition components directed along the
ccaxis, i.e., the matrix elements of x:
'l [(
2h ** \\ co,,o
  
2h ** \\ co,,o o> co M 4 co
n
(36.16)
We have made use of the Hermitian nature of matrix elements
expressed by the relationship (34.15). In other words, we have put
%,  2 in place of x Qn x n Q.
Now, by performing a simple algebraical transformation and
introducing the electric field E itself instead of its amplitude 7 ,
we have:
^ yt 2co M0 e 2 a? OM  2 v /Q17\
d = L *') E ' (36J7)
n
The dispersion formula. Let us consider the polarization of a medium
P = N&, where N is the number of atoms in unit volume.* The electric
induction is related to the electric field and polarization by the relation
ship (16.23) which, in the given case, is of the form
D = E + 47UP = l + JTyE  < 36  18 )
n
But D sE from the definition of dielectric constant, so that
c = i + I^JT 2 y^i a . (36.19)
h WMO
We note that this expression is correct only when the frequency
of the incident radiation is not close to one of the natural frequencies
of the atom <o,, . Otherwise the denominator in (36.19), and cor
respondingly in all the previous equations up to (36.10), can become
zero ; in any case, it becomes small. But then the perturbation caused
by the field is large, and it cannot be regarded as weak, utilizing
the expansion ^^^(oj+^d) and neglecting  ^  2 . Physically, this
means that if the frequency of the incident light is close to the fre
quency of one of the absorption lines (or, what is just the same,
the emission lines), we must take into account the damping in the
amplitude of the excited atomic states due to radiation. In other
* Such additivity of dipole moments is true only for gases.
See. 36J QUANTUM THEORY OF DISPKRSION 383
words, the amplitudes of the excited atomic states must not be taken
with a purely imaginary time dependence, but of the form (28.14):
> =  < (36.20)
The quantum evaluation of F for radiation damping is rather
complicated, and we shall not deal with it.
A comparison of classical and quantum dispersion formulae. We
shall now go over to a comparison of the quantum formula of dispersion
(36.19) and the classical formula. We shall write the latter for the case
when there are many types of oscillation whose frequencies are to no
(instead of a single frequency co ). We shall separate the total number
of oscillators N into parts corresponding to each separate frequency co no :
N =ZN n . (36.21)
n
If we introduce the relative fractions of each oscillation by the formula
fn^%', (36.22)
then it is obvious that
1 (36.23)
Let us suppose that the frequency of the incident radiation is
not close to any one of the natural frequencies CO MO . Then the classical
dispersion formula is generalized to the case of many frequencies
in the following way:
'
e = 1 + 7 z fn a . (36.24)
1 vn. ^J /.\2 ,.v2 \ '
m
Comparing it with the quantum formula of dispersion (36.19),
we see that both formulae become identical if we put
 2 . (36.25)
But to make this equation meaningful we must, in accordance
with (36.23), impose upon the righthand side of (36.25) the condition
= 1  < 36  26 >
Let us prove that this is actually what occurs. To do so we write
the commutation relation between p x and x [see (29.3)]
. (36.27)
384: QUANTUM MECHANICS [Part HI
We multiply it by ^5 on the left and ^ on the right (we omit the
upper index 0) and integrate over the whole volume. We can take
advantage of the normalization condition on the righthand side
of the equality, i.e., of the fact that ( <p 1 2 ^ ^ ^ 1:
J ftpxxhd V  J ^ Q xp^d V  \ . (36.28)
We expand the products x^ Q and ^x in an eigenf unction series
according to the general formula (30.8):
(36.29)
The expansion coefficients are determined from (30.11):
a n  J <K#<M V ; a: =  foxtynd V . (36.30)
In other words, they are equal to the matrix elements a? . Now,
substituting the expansion (36.29) into (36.28), we obtain
* . (36.31)
But this expression contains the momentum matrix elements, which
can be replaced by coordinate matrix elements by equation (34.21):
J fyoPxfynd V ^ (p x ) nQ = i
J fynP*fyod V = (p x ) Q n
After this substitution, equality (36.31) can be easily reduced to
the required form (36.26) if we take advantage of the fact that x no =xo n
and o) = 6) .
We note that the oscillator fractions / (they are also called ' 'oscilla
tor forces") are proportional to the same matrix elements as are involved
in the probabilities of radiation or absorption of the appropriate
quanta. Therefore, the dispersion properties of a substance may be
associated with the intensity of the spectral lines emitted by it.
Incoherent scattering. In addition to the dipole moment d deter
mined by equation (36.11), we can also calculate the transition
moments corresponding to radiation with a frequency which is less
than that of the incident light. In other words, we can calculate the
intensity of light scattering with a change of frequency. Such scattering
is termed incoherent.
A very important case is when the radiation energy, which remains
in the substance upon incoherent scattering, contributes to exciting
See. 37] QUANTUM THEORY OF SCATTERING 385
the oscillatory motion of the molecules. This phenomenon was dis
covered by L. 1. Mandelshtam together with G. S. Landsberg and,
independently, by Raman. It is frequently accompanied by the exci
tation of oscillations which do not manifest themselves in the direct
absorption of quanta as a result of the appropriate selection rules for
molecular oscillations. In this case, incoherent scattering yields im
portant information concerning the molecular structure of substances.
Sec. 37. Quantum Theory of Scattering
The effective crosssection concept in quantum theory. The concept
of an effective scattering cross section of particles, which was denned
in Sec. 6 in terms of classical mechanics, is directly extended to
quantum mechanics. Indeed, the differential effective scattering
crosssection of the particles inside a given solid angle is the ratio
of the number of scattered particles in this element of angle to the
flux density of the incident particles. Since flux and flux density
can be defined quantummechanically, the effective crosssection has
the same sense in quantum theory as it has in classical theory.
In practice, however, it is very difficult to calculate the effective
crosssection. Therefore, we shall consider certain special cases in
which the solution to the problem is comparatively simple.
The Born approximation. Let us suppose that a particle witli
energy $ is scattered in a given potential field U. We shall first
consider the case of $ > U. Then, the change in the wave vector
of the particle in the field is of the order
 U)
If the dimensions of the region in which the field acts are of the
order a then the total phase change of the wave function in the
scattering field is estimated as
L_ JZ5L
h '
This quantity must be considerably smaller than unity in order
that the perturbation produced by the field may be regarded as
weak. In the case when l]^>$, the wave number is estimated as
!/_..._ ^> /_ * . it follows then that the criterion of smallness
y ti \ ti . 
for a phase change is^y <^ 1 (upper estimate).
Under these conditions the action of the field U must be regarded
as a weak perturbation imposed upon the wave function.
We shall proceed from the general formula (34.29) for the transition
probability. Let the initial momentum of the incident particle equal p
25  0060
386 QUANTUM MBCHANirs [Part III
prior to scattering in a eentreofinass system (see Sec. 6), and p'
after scattering. We consider the scattering to be elastic, so that
p p f . To a zero approximation we choose the wave functions <]/> (p)
and <J/ 0) (p') in the form of plane waves, which corresponds to free
motion. We write them as
4,(o> (p) = y e ^T . </o). (p/) = ^y e  * . (37Bl)
These functions are normalized to unity in the volume V (which,
of course, falls out of the final result). The approximation (37.1)
for </> (p), </>* (p') is called a Born approximation.
The function <J^> (p) corresponds to a flux density ^ ; this is im
mediately seen from (24.20):
V eeVe = ~  . (37.2)
* V \ ) mV V v '
From (37.1), the matrix element for the transition probability that
appears in (34.29) is
~ 7> ~~ U(t)dV. (37.3)
In order to find the scattering probability, we must multiply the
square modulus of (37.3) by~ , by the number of finite states z ($)
in a unit energy interval, and by the element of solid angle di.
This number can be determined directly from (25.25) if we take
into account that the fraction of the states corresponding to a solid
angle element dii is^ .
Therefore,
(37.4)
The differential effective scattering crosssection inside a solidangle
element d ii is equal to the scattering probability in unit time [defined
from (34.29)], divided by the incident flux^ = 4 p^ Therefore,
V
J
(p
U(t)dV
2
m*
dSl . (37.5)
The integral appearing here is the matrix element calculated for
two functions <j/> (p), <J/>* (p') with normalization over unit volume,
F = l. Therefore, introducing k^= J, k'= , we A\Tite
Sen. 37 J QUANTUM THEORY OF SCATTERING 387
Uw = je< (k  k '> ' U (r) d V, (37.6)
'kk'
(37.7)
Scattering by a central field. Simplifications appear in expression
(37.5) if the field U is central, i.e., if it depends only upon r. Let
us calculate C7 k k' for this case. In defining the polar angle & we choose
the direction of the vector k k' as the polar axis. Then
. (37.8)
Bearing in mind that sin fr d$= d cos ft, we can integrate with
respect to & immediately, obtaining
! , p /kk'r_ _ik k'r
TT rt I O J T J I \ I " J ^^ ^
11..* V 7"f I 4** // "f* / / I / 1 I 
l_x IrV *j/tl/ Iv/l/l/ll it 1/1
J v ' \  k k'  r
o
oo
= Tb 4 Vr fr 17 (r) sin ( I k  k' I r) d r . (37.9)
kkJ
As we have already said, k~k'. Therefore, the vector difference
is easily expressed in terms of the deflection angle for the particle :
k k'1 2 ^ 2fc 2  2(kk') = 2A; 2 (1 cos 6) = 4 A; 2 sin 2 ^ . (37.10)
1 ' Jt
This can also be seen from a geometrical construction. We have
oo
 ~ CrU(r)am(2krsm~\dr. (37.11)
Uw= *sin. J \ 2/
o
Thus, a calculation of U^^ reduces to calculation of a single integral
(37.11).
Rutherford's formula. For the case of a Coulomb field, U =  .
The integral E/kk' is found in the following artificial manner. We
00
define the integral (sin x dx thus
o
oo
lim I e~~ a * sin x dx = lim ~ 2 ~~i ~ 1
<,~>l)J ..O^i 1
25*
388 QUANTUM MECHANICS [Part III
Then
CO
f 1
J ~~ a
and, finally,
tfkk = 2nZe ~ I sin(2A;rsm^)dr = TC " P " (37.12)
Substituting this in (37.7), we obtain a final expression for the
differential effective scattering crosssection:
4 m 2 y 4 sin 4  
where we have taken advantage of the fact that p~hk~mv. This
result curiously agrees with the precise classical Rutherford formula
(6.19).
It turns out that the result (37.13) is also obtained from a precise
solution of the wave equation for the case of a Coulomb field. Thus,
Rutherford's formula is extended to quantum mechanics unchanged.
The Born approximation in the theory of scattering by a Coulomb
field can be regarded as a series expansion in square powers of the
charge, or, more precisely, Ze 2 . But since the precise formula does
not involve powers higher than (Ze 2 ) 2 , the result of the Born approxi
mation coincided with the precise result.
We shall now estimate the limits of applicability of the method
under consideration for the Coulomb field. To do this, we make use
of the first criterion established at the beginning of the section for
the applicability of this method. Since the product Ua in this case
is equal to Ze 2 , we arrive at the following condition:
(37.14)
The quantity e 2 /hc = 1/137. Therefore, we write (37.14) otherwise
thus:
wH 1  < 37  15 )
But Z~90 for heavy elements, so that (37.15) is not satisfied in
general. Of course, Rutherford's formula is applicable to nonrela
tivistic particles in this case too, because it is exact ; but in calculating
a correction, for example, arising from a distortion of the nuclear
field by the field of atomic electrons, the Born approximation yields
an incorrect result.
Sec. 37J QUANTUM THEORY OF SCATTERING 389
With the condition (37.15), formula (37.13) is applicable also to the
scattering of relativistic particles at small angles provided m is re
placed by rJ==r .
The collision parameter (aiming distance) and angular momentum.
The Born approximation cannot be used when large forces act upon
a particle, even if they are concentrated in a small region. First of all
let us define what is meant by a "small" region.
It is convenient here to compare the classical aiming distance of p
(see Sec. 6) with the angularmomentum eigenvalue in quantum
mechanics. For large angularmomentum eigenvalues, when the quasi
classical approximation is applicable, we can give the following esti
mate :
hl~mvp. (37.16)
Whence
p~ hl  X/ , (37.17)
r m 2r> ' v '
Avhere X is the de Broglie wavelength.
It can be seen from here that to a change in the angular momentum
by unity there corresponds an increment of  , in the aiming
distance. Accordingly, the smallest collision parameter is given by
/==0 and p ^   . Here the particle is scattered iu the sstate.
Let us consider the case when the radius of action of the forces is
less than  . Then a particle with an angular momentum other
than zero hardly at all experiences scattering. We have shown [see
(31.12)] that the wave function for a particle with angular momentum
I becomes zero, like r l , at the origin. Therefore, the probability of
finding a particle with / > in the region of action of the forces is
very small if the radius of action of the forces is much less than y .
ZTC
Separating the wave function with zero angular momentum. Let
iis take the term, corresponding to = 0, out of the wave function.
To do this, it is necessary to expand the function (37.1), i.e., a plane
wave, in a series of eigenfunctions of the operator M 2 . The function
corresponding to 1 = is especially simple: it does not depend upon
the angle. Indeed, the operator M 2 involves only angular derivatives.
Operating upon a function which does not depend on the angle is
equivalent to multiplying the function by zero. Normalizing the angu
lar function of the sstate to unity, we find
f
then J cp  2 dQ =
390 QUANTUM MECHANICS [Part HI
The expansion coefficient for this function is, according to the gen
eral formula (30.11),
c(r) =
_
^ fe' fercos * sin db = 2 1/jT ^L^L . (37.18)
V47TFJ r F kr
c (r) satisfies the radial wave equation for a free particle (31.5),
if we put C7 = 0, I in it. In actual fact, equation (31.7), which is
obtained from (31.5) by substituting <I* = y wnen & = 0, 7 = 0, has
the solution _
__ . <\/2m&
X sm ^ r
But   J^ ~ k, so that the function ^=^rc (r).
c (r) tends to a finite limit when r 0. This corresponds to the
boundary condition for a radial ^function at the coordinate origin.
Let there now be, close to the origin, a scattering field U (r), which
diminishes so rapidly with distance that U (r) = when r ~ . Then,
in its radial dependence, the s state wave function satisfies, as before,
equations (31. 5) (31.7) for T>^ 9 because no forces act upon the
particle in this region. The solution to (31.5), which is more general
than (37.18), is
. (37.19)
where 8 is some phase shift depending upon the definite form of the
potential U (r). Naturally, the solution (31.19) cannot be extended
to r < 5 , because the particle in this region is no longer free from the
action of forces.
We shall show that the effective scattering crosssection is expressed
in terms of S. An example of determining S is given in exercises
2 and 3.
Determining the scattered wave. Let us suppose that, in some way,
an exact solution </ of the wave equation (31.12) has become known
in a given scattering field U (r). This solution must satisfy the same
conditions at infinity as a plane wave, because these conditions corre
spond to the particle scattering problem. We represent the function
</ in the form of the sum
*' = *(P) + *** (37.20)
Sec. 37] QUANTUM THEORY OF SCATTERING 391
This equation separates the function tb (p), which corresponds to
the incident wave, from the complete solution of the wave equation.
The second term ^scat=^' ^ (p) describes the scattered particles.
If we now expand ^' and <J> (p) in a series of eigenfunctions of the
square of the angular momentum, it turns out that a shortrange force
field distorts only the term <]/, which corresponds to 1 = 0. All the
remaining terms of both functions are the same. As a result, when
r >  we obtain
47t
i r//\ / \T nl/~rc [Aain(kr + B) amkr] * /orroix
<kcat  [c' (r) c (r)] 9 = 2 / T _ ^ J ^j= . (37.21)
But for large r the scattered particles can move only away from the
scatterer. This means that ^at involves only the function and
does not contain the function . Indeed, if we put i = in
r * r
(24.20), then the flux j will acquire a positive sign, while if we substi
tute '  the flux will be negative, i.e., incoming.
We write d> scat in complex form:
<J, scat =    [A (e''<*' + *>  e '<*' )  (e ikf  e~ ik ')] . (37.22)
2i v V ' kr
To exclude the incoming wave e~ ikr , we put 4e~' 8 ~l, whence
4e' 5 , (37.23)
(e2i8  1)eI ' kr  (37  24)
The effective scattering crosssection. From (24.20), the flux density
of the scattered particles at infinity is
7 = Ti/irr I e'  1 1 . (37.25)
1 4Vk 2 r 2 ' ' m v ;
The total effective scattering crosssection or is equal to the whole
flux 4Tcr 2 ; divided by the flux density of the incident particles
v _ hk
'V ~~ ~mV ''
(jz=^.e 2 ~ 1 2 . (37.26)
K
Passing to real quantities we write
 e 2i _ 1 1 2 = ( e 2i_ i) (ea 1) = 2  2 cos 2S = 4 sin 2 8,
so that
(37.27)
392 QUANTUM MECHANICS [Part III
It can be seen from here that the greatest value of cr is for the scatter
ing of particles in the sstate  . This formula has many applications
in nuclear physics since nuclear forces are large and shortrange.
Since scattered particles occur in the sstate, their distribution in
a centreofmass system is isotropic, i.e., it does not depend upon the
scattering angle. This agrees with the statement made at the end of
Sec. 6.
Exercises
I) Find the effective scattering crosssection of fast particles by hydrogen
atoms in the ground state.
The wave function for the ground state of a hydrogen atom with n = 1,
I  is
because, of the polynomial x there remains only the first term, which does
not depend upon , while \/2e ~ ! The coefficient It is found from the nor
malization condition
whence
RRJT
The potential interaction energy of the charge c with the atom is
The first term of the integral f/kk' was found in the text. It is
Wo integrate the second term in the following way:
C J G dV j ""fT^rT" "
 2 f 2 > i<klOr' , c '( k  k ')(rr')
J J  r r' 
In the last integral it is necessary to take the origin at the point r', so
that it reduces to the same form as (37.12):
J f
Hence,
Sec. 37 J QUANTUM THEORY OF SCATTERING 393
The quantity inside tho brackets is callod the screening factor. Evaluating
it in the same manner as (37.8), we obtain
f / i(klOr" r, 2rr f , L , . 0\
The integral reduces to the form
OO CO
J X Sm ** e C * l ~ ~~ 0V J Sm " ** P/ '* ~~ 0V <
2nb
Here, a = 2 k sin , 6 = ~ , so that the screening factor is
1 / w e 2 \ 2 r
~~ r:\li* r /"" 7 .

~
07. sin 2wrS
  ^ sin o~ ' h* I
., . o 2we 2 M a "" "" I . h* k 2 ' 7" " 01
A* sin
1 . .
hV \* ,' 2 I'" '
It was assumed in tho last transformation that tho scattered particle is an.
electron. Then, strictly speaking, we should have formed a function which
is antisymmetrieal together with the function of the atomic electron; this
we did not do. The final formula for the effective scattering crosssection
differs from (37.13) by the square of the screening factor. We note that this
factor is correctly obtained only in the Born approximation, hi contrast to
Rutherford's formula (37.13), which is exact.
For 00, the effective crosssection turns out to bo finite, because 0^0
corresponds to large aiming distances, when tho nuclear charge is screened
by the charge of an electron.
2) Calculate the effective scattering crosssection for a particle by an im
permeable sphere of radius a, which is very much less than 7: ~ , .
Z TC K
In accordance with (25.1), the wave function at the surface of tho imper
meable sphere becomes zero. Hence, the solution (37.19) has tho form
k (r a)
kr
From this, 8 == ka, while
But ka <^ 1 from the conditions, so that sin ka ~ ka.
Finally, a 4 TT a 2 , i.e., the effective scattering radius is twice tho radius
of the sphere. In classical theory or = jra 2 (see exercise 1, Sec. 6).
3) Examine the scattering of particles with energy & in the sstate by a
spherical potential well of constant depth  U Q \ and radius a; consider that
there exist the following relations:
394 QUANTUM MECHANICS [Part 111
(see Sec. 25). For e > 0, there exists in the well a bound state of a particle
of energy <f close to the upper boundary of the well. Unlike Fig. 38, it is
assumed here that U becomes zero for r > a. Express the cross sect ion in
terms of the energy level ^ .
The conjugation condition for the wave functions with r = a is of tho form
x cos xa
sin (fcof 8)
where x = ~ r ' ~ , k = = Neglecting ka, wo obtain
fi n
an expression for the effective crosssection:
4*
a = TV
~k* (1 + cot 8 8) k 2 + x 2 cot 2 xa
In accordance with the conditions imposed upon E/ \U \ wo have, approx
imately,
cot xa   .
4
From (25.41) the condition for finding the level *? is of the form
x cot x a ~ x
Supposing that
e , we obtain
^0
assumption concerning the order of smallnoss
the cross section we finally have
4*
This confirms the
, since V e ^ e For
__
where k  ^^ ? ' We note that the formula obtained also holds
for e < 0, when in actuality there is no level at all in the well. In this case
we talk about a "virtual" level. The straight line in Fig. 39 intersects the
first halfcycle of the sinusoid just before x a == ~ .
A similar case occurs when neutrons are scattered by protons with anti
parallel particle spins.
Sec. 38. Tho Belativistic Wave Equation for an Electron
The equation for a spinless particle. Schrodinger's equation (24.11)
is formed on the basis of the nonrelativistic relationship between
energy and momentum
Therefore, it can be applied only to electrons whose velocity is consid
erably less than that of light, and whose kinetic energy is considerably
less than the rest energy:
Sec. 38J THE BELAT1VISTIC WAVE EQUATION 395
T= 2
Immediately after Schrodinger obtained the nonrelativistic equa
tion, the first attempts were made to build a relativistic wave equation
(Fock, Klein, Gordon). In formula (21.30)
 e?) 2 = c 2 (p  ~ A)" +
w a
\ m c /
~ ~dt was substituted in place of Jff, as is usual in quantum mechan
ics, and in place of p, the operator 4 V. In this way a wave equation
was obtained in relativistically invariant form
8
m 2 c 4 ^, (38.1)
which equation, however, is not applicable to electrons. The fact of
the matter is that equation (38.1) does not take into account the spin
of the electron, because it involves only a single wave function. Yet
in Sec. 32 we saw that a particle with spin 1/2 must be described by
at least two wave functions. These two functions could be introduced
into the nonrelativistic equation purely formally, assuming that each
of them satisfies it. But the interaction of spin and orbit is a relativ
istic effect; therefore, a correct equation for fast electrons must take
it into account automatically, without any additional hypotheses
concerning spin magnetic moment. This equation must involve op
erators which act upon the spin degree of freedom.
The inapplicability of equation (38.1) to the electron was very
quickly seen: the fine structure of the levels of the hydrogen atom
obtained from this equation was incorrect. A nonspin equation cannot
explain, first of all, the number of splitting components; this is deci
sively against it.
Charged particles without spin mesons take part in nuclear
interactions. Equation (38.1) can be applied to them, at least if it is
shown that such mesons can be regarded, to some sort of approxi
mation, separately from protons and neutrons.
But for electrons one has to form a relativistic wave equation that
takes spin into account. Such an equation was obtained by Dirac.
The Dirac equation. Following the line of Dirac 's argument, we
begin with the equation for a free electron. The starting relationship is
ff = Vc 2 (pi + p$ + pi) + m 2 c* . (38.2)
390 QUANTUM MECHANICS [Part III
Instead of and p we must substitute the derivatives ^j and
j l> G t
.V. However, to do this it is necessary to define the meaning of the
square root of an operator. Dirac supposed that, in the operator sense,
a root is equal to an expression like
Vc 2 (p + * + p z ) + w 2 c 4  c (* x p x + 5i y p Y + a z p z )
(38.3)
where <x x , a y , a*, and p act on the internal degrees of freedom of the
electron such as, for example, the spin degree of freedom.
Let us square both sides of the equation and attempt to choose the
operators a*, a y , a*, and p in such a way as to obtain an identity, i.e.,
so as to eliminate terms of the type p x p y , . . . , me 3 p x , . . . :
c 2 (pi + P 2 Y + pi) + m*c*  c 2 (alpl + a JfiJ + Upl) + P 8 w a c* +
+ r, a (a x a y + S. y S. x )p x p y  r, 2 (a. v a^  Si z S. x )p x p s +
+ c 2 (a y a^ f $.zS. y )p y pz + wc 3 (a. x p + Pa. x ) /)* 4
h ?wc s (a y p f Pay) p y + mc 3 (a^p + $$.*}$?.
Hence, the operators must be subject to the conditions
 . X . Z \ aa. v =
= a^p + PCX^^ a y p + pa y = a ? p + pdt^ = . (38.4)
These operator equalities greatly resemble the spin operator relation
ships (32.10), (32.11). It can already be seen from this that the oper
ators S. x , a y , a*, and p at least act upon the spin degree of freedom of
an electron. To the accuracy of the factor 1/4, the relations (38.4) agree
with (34.11) and (32.13) for S x , 5 y , and S z .
But the operators a and o are not identical. This can easily be seen
by proceeding from the opposite : assume that . X = G X , 6c y ^5 y , S. Z = S Z .
In order to obtain the wave equation we must equate the righthand
rot
side of (38.3) to  ""=T* ^ ow ^ us P er f rm an inversion of the
coordinate system. All momentum components will change sign so
that the sign of p z will also change. But the operator 6c^ in front of p z ,
if it equals S z , should not interchange the wavefunction components.
Therefore, the equality between the left and righthand sides of the
wave equation breaks down when the coordinate system is inverted;
but this should not be. Therefore, a 7^ a.
The necessity for a fourcomponent wave function. We shall be
confronted with the same difficulty if we consider that the operators
a*, a y , a* act upon the same wave functions as those appearing with
Sec. 38] THE KELATIVJSTIC WAVE EQUATION 397
l
 ~~dT n ^ e ^ ier S ^ e f the equation. Therefore, we must assume
that coordinate differentiation, on the one hand, and time differentia
tion, on the other, are applied to different pairs of functions, of which
one pair changes sign in inversion of the coordinate system while the
other does not. This is sufficient to ensure invariance of the equation
with respect to inversion.
Thus, we shall say that the wave functions depend not only upon
the spin variable a,, but also upon some other internal variable p,
which also takes on two values.
Let us define the operators which are completely analogous to spin
operators and which act upon the variable a and upon the variable p.
Bearing in mind that the factor 1/2, with which the spin operators
were multiplied by in Sec. 32, is no longer needed, we shall write
6 1? o 2 , 6 3 and, correspondingly, p x , p 2 , and p 3 for the variable p. These
operate upon the wave function analogously:
<*2
\y v~, p;/ v < y i l > pyy
f' Ji n o\\
(38.5)
Pi
^ 3 \if '2n~\ ib C '2^)' (38.6)
From formulae (32.10) and (32.11) we find the basic relations for the
operators a and p :
iS l \ 1
f
J
and similarly for p.
All the operators o are commutative with all the operators p
because they operate upon different variables. In order that the
operators a*, a y , SL Z together should satisfy the "anticommutative"
relations with the operator (, as in the third line of (38.4), we arrange
that all three components of a are proportional to one operator of the
components of p, for example, p x and p is simply p 3 . We notice that
p x interchanges the functions while p 3 does not. To have the operators
a*, a y , and SL Z anticommutative, as in the second line of (38.4), we
put them proportional to a v o 2 , 6 3 , respectively. Thus,
a^^p^a, P = p 3 . (38.8)
398
QUANTUM MECHANICS
[Part III
It is obvious that, as a result of the commutative nature of p and a
and of the definition of the operators (38.5) and (38.6), all the opera
tors formed in (38.8) satisfy the conditions (38.4).
Thus, the wave function in the Dirac equation has four components,
according to the number of values for o and p (a 1, 2; p = l, 2).
For convenience in future we shall number the symbols a, p from one
to four, putting
It is convenient to write the functions ty l9 i]; 2 , ^ 3 , ^4 i n columns, as in
Sec. 32. Now using (38.5) and (38.6), we find out how the operators
and p act upon the fourcomponent wave function
* (1,1)1
a y , a*
(2, 2)
= Pi
(2, 2)
*4
I;
V,
(38.9)
The choice of operators (38.8) and (38.9) is not unique. It is possible
to form other operators with the same properties. For example, one
could have chosen p 2 instead of p x . We shall examine below (exercise 4)
the implications of this fact.
The Dirac equation in expanded form. Summarizing, according to
(38.2), (38.3) and (38.8) Dirac's equation can be written as
(38.10)
In accordance with (38.9), this equation must be understood as a
system of four equations, which we shall write explicitly, first of all
T p  I
replacing  ~~^7^J &$ (because fy is proportional to the factor
iEt \
e'~ * ) :
= c (p x
= c (p x
= C (P*
= C (p x
4 p z
(38.11)
h
As usual, here p x , p y , p z are the components of the vector V, i.e.,
. ^^7 . The system of equations (38.11) is applied to a
7 "FT
~~ )""
Seo. 38]
THE BELATIVISTIC WAVJ5 EQUATION
300
free electron since it does not involve the scalar potential and compo
nents of the vector potential. Therefore, the coordinate dependence of
the wave function is determined by the factor
so that the whole group of four fy has the form
JH
(38.12)
where the amplitudes a l9 a 2 , 3 , and a 4 do not depend upon coordinates.
The action of the operators p X9 p y , p z on this group of four fy leads
simply to a multiplication of its components by p X9 p Y9 p x . Conse
quently, the differential equation (38.10) for a free electron leads
to the algebraic equation
+ wc 2 pa, (38.13)
where by a is understood the whole column
a =
Here, operator properties are preserved only by a*, a y , a*, and p,
which rearrange the amplitudes a in the same way as the functions <{/.
In other words, the amplitudes depend upon the internal variables a
and p.
Energy eigenvalues. We apply the operator
$ = c 2 (ocp) + mc^ p (38.14)
to both sides of equality (38.13). As a result of the anticommutation
properties of the Dirac operators (38.4), only their squares, equal to
unity, remain on the right. Hence, the following equation will result :
W 2 c 4 a.
(38.16)
Here, the components of a are no longer interchanged because there
are numbers in front of a, and not operators. All four equations (38.15)
have the same form <? 2 a x = c 2 p 2 a t + w 2 c 2 a l9 etc. For these equalities
to be satisfied we must subject the energy to the usual relativistic
relationship, i.e., we cancel a x :
(38.16)
400 QUANTUM MT3PHANICS [Part III
The last equality refers simply to magnitudes. Thus the energy
eigenvalue for a free electron, determined from the Dirae equation, is
cM 7 ^ 2 . (38. 1 7)
The two signs in the equality correspond to the internal degree
of freedom which an electron possesses in addition to spin. Only the
plus sign is taken in classical mechanics, since free electrons do not
have negative energy. The square root in (38.17) is not less than
me 2 in absolute value, so that a region of width 2 me 2 exists in which
the energy cannot occur. But all the quantities in classical equations
vary continuously; therefore, once the energy has been defined with
a positive sign, it cannot jump across this forbidden region of width
2 me 2 , and remains positive all the time. In other words, energy which
is defined as positive in the initial conditions remains positive from the
equations of motion.
Negative kinetic energies in the Dirae theory. In quantum theory it
is shown that discontinuous transitions, too, are possible between
different states. For example, an electron with energy greater than
we 2 could emit a light quantum and remain with an energy less than
me 2 . But such electrons with negative energy and mass are not
observed in nature. Their properties would be very strange: upon
radiating light, they would each time reduce their energy and, as it
were, drop into a state with ? = oo. All the electrons in the universe
would rather quickly fall into this state ; but, as we see, this has not
happened !
Thus, Dirac's equation admits of the possibility of states, which
cannot simply be excluded, because electrons may transfer to them
from other observable states. But, on the other hand, there are no
electrons in nature with negative energy & Vm 2 c 4 +c 2 p 2 . At the
same time, the Dirae equation describes quite correctly a great assem
blage of electron properties : as we shall soon see, it yields a relationship
between the spin and magnetic moment of an electron that agrees with
experiment, it leads to an accurate formula for the finestructure levels
of the hydrogen atom, etc. In addition, mathematical investigations
show that there is 110 essentially different relativistically invariant
wave equation for a particle with spin one half and mass differing from
zero. Therefore, one should not simply reject the Dirae equation: it
is better to attempt to supplement it with some kind of hypothesis.
Vacuum in the Dirae theory. Dirae suggested that vacuum should be
redefined. Earlier, vacuum was understood as being a state of matter
in which there are no charges, say electrons. A vacuum must now be
called that state in which all negative energy levels are occupied by
electrons. That this redefinition is not verbal but has a physical mean
ing will be seen very soon from what follows.
Sec. 38] THE RELATIVISTIO WAVE EQUATION 401
If all the negative energy levels are occupied, then, in accordance
with the Pauli principle, no electrons can transfer to them from positive
energy states. Thus, the Pauli principle is necessary for relativistic
quantum theory to be able, in general, to describe the properties of
electrons. This is the basic reason why the Pauli principle is necessary
as an element of quantum mechanics. In order to avoid misunder
standing we shall give a somewhat fuller definition of a "vacuum"
in field theory, which has a rather different meaning from that in
experimental physics. In field theory, vacuum signifies the ground
state of the field; for an electromagnetic field, for example, it is the
state of the field in which there are no quanta. We saw in Sec. 27 that
this state is endowed with observable physical properties.
In exactly the same way, if all the negative energy states are occu
pied, then all the remaining electrons can no longer reduce their energy
by making a transition to negative states. When there are no electrons
with positive energy, electrons can no longer reduce their energy in
any way if all the negative energy states are occupied. This explains
the definition of vacuum as a ground state.
Pair production. All observed phenomena occur, so to speak, on the
background of a state in which the negative kineticenergy levels are
filled.
However, this "background" can manifest its existence in a real
physical process. Close to a nucleus, a quantum with energy greater
than me* is capable of effecting the emission of an electron from a nega
tiveenergy to a positiveenergy state. Proximity to the nucleus is
necessary so as to satisfy the law of conservation of momentum. For
the proof of this simple statement, see exercise 1.
But after an electron has been removed from a negative energy
state there will remain a "hole," i.e., an unoccupied level. In an elec
tric field, electrons with negative mass (mass has the same sign as
energy) do not move against the field, towards the anode, but along
the field, to the cathode, against the applied force. And with them
moves the hole which thus behaves like an electron of positive charge
and mass.
Experiment will show that as a result of the ejection of an electron
from a negative energy state, two charges have appeared: negative
and positive. Such a positive electron, or positron, was discovered by
Anderson after Dirac had formulated his theory of the background.
The attitude towards the Dirac equation was somewhat suspicious
before the discovery of the positron, while the idea of background was
considered farfetched and intended only to hide the defects of the
theory.
In actual fact Dirac 's theory is an unusual example of scientific
foresight. The discovery of the "antiproton" (i.e., a proton with
negative charge) by Segrfc once again confirmed the generality of
Dirac J s conceptions concerning particles with spin 1/2.
26  0060
402 QUANTUM MECHANICS [Part III
Pair annihilation. When a positron and electron meet they can
annihilate each other if the electron transfers to an unoccupied level
belonging to a negative energy state. Its energy will be imparted to
electromagnetic radiation in the form of two or three quanta. A single
quantum cannot result when annihilation occurs in free space because,
in this case, momentum is not conserved, in the same way that a single
quantum cannot form a pair in free space. Singlequantum annihi
lation, too, is possible in a nuclear field.
The Dirac equation and quantum electrodynamics. Modern quantum
electrodynamics is based upon the quantum theory of the electro
magnetic field (Sec, 27) and the Dirac electron theory, with account
taken of direct and reverse transitions from negative energy to posi
tive energy. It actually describes an electron and a positron in a com
pletely symmetrical way: the nonsymmetry of the charge that has
appeared in our terminology as a result of the fact that the positron
was defined as a "hole," is only apparent. The background of elec
trons with negative energy may be, as it were, subtracted from the
equations without altering the physical content of the theory, and so
the equations become symmetrical with respect to the sign of the
charge. The concept of particles and antiparticles is extended also to
particles without spin (for example, TC+ and 7r"mesons).
It is characteristic that the relativistic quantum theory describes
change in the number of particles: electrodynamics treats of ab
sorption and emission of quanta, electron theory is concerned with
the creation and annihilation of pairs.
The electronpositronphoton field. Electrons together with a field
form a sort of unified electronpositronelectromagnetic field. Insofar
as interaction exists between electrons and positrons, on the one
hand, and quanta, on the other, the division of a unified field into
charges and quanta becomes, in a certain sense, artificial and, in any
case, approximate. The strength of interaction between field and
charge is determined by the dimensionless parameter
e 2 _ 1
he ~" 137 '
Since this parameter is a rather small number, the approximation
arising from the separated charges and field seems satisfactory.
As was mentioned in Sec. 27, the theory still has certain difficulties.
The most difficult problem even the approach to which is not
known consists in an explanation of the number r= : since it is an
abstract number, the theory shbuld, in principle, derive it from certain
general physical principles. But these principles have not yet been
formulated.
Nevertheless, all problems in which quantum electrodynamics is
used in calculating experimental quantities have complete and unique
Sec. 38] THE BELATIVISTIC WAVE EQUATION 403
solutions. Therefore, despite certain imperfections, the quantum theo
ry of the electromagnetic field possesses the essential features of a
correct theory: agreement with experiment and a very specific mecha
nism for calculation.
The quantum theory for other fields is quite a different matter.
The only other comparatively satisfactory situation is in the theory of
the field responsible for the beta disintegration of nuclei and other
socalled "weak interactions," like TupLmeson decay, etc. (see below).
As regards the theory of nuclear forces, all that is known from a
series of unsuccessful attempts is what form the theory cannot have.
However, there are many experimental facts that permit of a con
clusion concerning the physical nature of nuclear forces. These forces
are undoubtedly related, at least in part, with the socalled 7cmesons
particles with mass 273 times that of the electron mass. These mesons
play the part of quanta in the field of nuclear forces. But they interact
with nucleons (i.e., with protons and neutrons) so intensely that it is
doubtful whether there is any sense in a separate consideration of
nucleons and 7rmesons as opposed to electrodynamics, where, to an
initial approximation, electrons and quanta may be considered sepa
rately.
In addition to remesons , there are heavy jfiTmesons, which dis
integrate into three, and sometimes only into two, 7umesons.
Upon decay, Timesons give fAmesons, which interact weakly with
nuclei. The role played by such weakly interacting particles in the
general scheme of nuclear forces is mysterious in the extreme.
An analysis of experimental data shows that the three basic types
of elementary interactions differ essentially as to their strength:
1 ) The strongest are nuclear interactions. These include, for example,
interactions between 7umesons and nucleons.
2) Electromagnetic interactions between quanta and charged ele
mentary particles are approximately one hundred times weaker than
nuclear interactions.
3) Interactions which are related to beta disintegration or, for
example, to the decay of heavy mesons of mass 960 electron masses
into 2 or 3 7umesons are weaker than nuclear interactions by a factor
of 10 11 .
Landau, and also Lee and Yang, have shown that the laws for weak
interactions cannot be invariant with respect to a simple transfor
mation from a righthanded to a lefthanded coordinate system (in
version). For the interaction to remain invariant, it is necessary,
simultaneously with inversion, to transfer from particles to antipar
ticles, i.e., from electrons to positrons, from protons to antiprotons,
from TC+ to 7c~mesons, etc.
Thus, the simple law of conservation of parity, which obtains for
nuclear and electromagnetic interactions, changes its form for weak
interactions. Starting with the principle of "combined parity,"
26*
404 QUANTUM MECHANICS [Part III
R. Feynman and R. GellMann (and, independently, R. Marshak
and Snolarshan) have succeeded in constructing a universal Hamil
tonian for all the weak interactions. The original form was suggested
by Fermi. It contains a new universal constant of the order of
10 49 ergcm 3 .
Classification o! elementary particles. It is rather difficult to define
exactly just what an "elementary" particle is. At the beginning of
this century, the atoms of elements were considered to be elementary
particles, since they were thought to be indivisible. Now we know
that atoms consist of electrons, protons, and neutrons, and we are
quite sure that these latter do not consist of still other particles of a
more "elementary" nature in the same sense as atoms do.
Several elementary particles transform into one another. In some
cases we can precalculate the laws of such transformations, for in
stance, for electrons, positrons, and photons, or, to a less extent, for
beta transformations; but we know very little about strong nuclear
interactions, which are undoubtedly also connected with some con
version of elementary particles.
To visualize this situation better let us consider the following proc
ess: a neutron emits a negative 7rmeson, and a proton absorbs it.
The neutron converts into a proton, and the proton into a neutron,
and the whole process of emission and reabsorption can be treated
as an "exchange" interaction. This sort of interaction is to some
extent analogous to electromagnetic interaction, where a photon is
emitted by one electron and absorbed by another, but unlike the elec
tromagnetic forces, we can describe the mechanism of nuclear forces
only in words. All attempts to do more have as yet failed.
Despite the lack of a theory of elementary particles, it is now possible
to bring them into some order. This classification is due to R. Gell
Mann.
We shall not describe the experimental proofs of the existence of all
the elementary particles listed below; such proof can be found else
where. Let us be satisfied in stating that their actual existence is quite
definite, in contrast to the "existence" of the ephemeral elementary
particles which disappear from the pages of scientific papers after a
careful investigation.
The GellMann classification is based primarily on particle inter
action. First, a particle exists which is capable of electromagnetic
interaction only. This is the photon, or the quantum of electromagnetic
radiation. Another group of particles is not capable of strong nuclear
interactions: the (jimeson, the electron, and the neutrino the so
called "leptons" (light particles). Still another group of particles,
consisting of TT and ^Tmesons, is capable of nuclear interactions. The
masses of these particles are intermediate between those of nucleons
and leptons. TT and ITmesons can appear and disappear in the nuclear
transformations of other highly energetic particles; no conservation
Sec. 38] THE RELATIVISTIC WAVE EQUATION 405
law for their number exists, but the charge conservation law is never
violated. The name baryon (heavy particles) has been given to a
fourth group of elementary particles. This group consists of stable
nucleons (neutron and proton), and unstable hyperons: A, JT^andS,
which transform spontaneously into nucleons. A conservation law
concerning the baryon number exists, and is as strong as the charge
conservation law.
All the particles listed above, except the photon and the 7umeson
have counterparts antiparticles. It is very important that a particle
need not to be electrically charged to be able to have an aiitiparticle,
as witness the antineutron. The process of particleantiparticle inter
action is one of annihilation ; thus, the antineutron and the neutron
are mutually annihilated, and in the process create 7rmesons. True
neutral particles are only those that do not possess antiparticles, or,
in other words, such that physically coincide with them in a trans
formation from the ordinary world into the "antiworld." This means
a mathematical transformation of all the wave equations interchang
ing positrons and electrons, negative antiprotons and protons, etc.
(some think that antinebulae actually exist in the universe).
If the physical laws governing the antiworld are the same as those
that govern our ordinary world, then the Hamiltonian of electro
magnetic interaction must be invariant under transformation from
one to the other. The sign of the charge evidently changes in such a
transformation, and the charge enters the Hamiltonian multiplied
by the amplitude of the electromagnetic field, the vector potential A,
so the latter must change its sign simultaneously. We conclude that
the amplitude of the photon is odd with respect to a transformation
from the ordinary world to the antiworld. The 7cmeson decays into
two photons; consequently its amplitude must be even.
Such parity of true neutral particles is one of their very important
characteristics.
We now pass to a iiontrivial point in the GellMann classification.
We first consider the decay of the electrically neutral A particle
A >7t+n or A>TC~ + P
(both are possible). The mean time of such decay is of the order of
10~ 10 sec, but the A particle itself was created in a nuclear collision
which lasted less than 10~ 21 sec. It appears rather puzzling that a
particle that can be created so quickly should disappear so slowly.
It seems to contradict the general reversibility of physical laws in
time. This was why the A particle came to be called "strange." The
only explanation is that both processes are of a totally different na
ture. The generation of a particle is due to a strong interaction, and
the decay, to a weak interaction. It is enough to suppose that the
A particle is always created from a nucleon accompanied by a
406 QUANTUM MECHANICS [Part III
jffmeson. This has been verified experimentally, though indirectly. Such
a process does not violate the baryon conservation law. If A and K
particles take part in the interaction simultaneously, it is strong,
and if only one of them transforms into something else, the interaction
is weak.
To distinguish between these two types of interaction, GellMann
introduced a new characteristic of mesons and of baryons their
"strangeness," S. It is defined as follows:
. 
e 2 ^ * ' 2 '
Here, Q is the charge of the baryon, e is the magnitude of the elemen
tary charge, n is the difference between the number of baryons and
antibaryons (1 for baryons, 1 for antibaryons, and for mesons),
and T* is the zcomponent of the isotopic spin (see Sec. 32). As all the
heavy particles take part in strong interactions, a definite value of T*
can be ascribed to each one of them, n must be taken equal to 1 for
each baryon. For nucleons, T* is + y, as they can have only two values
of charge: 1 and 0. A has no charge and can have only T g = 0. JT 1 has
three values of charge: 1, 0, 1, and equal values of T*. Lastly, S is
either neutral (r*=l/2) or negative (T* 1/2). Substituting these
values of charge and T* into the definition of strangeness, we find that
nucleons have /S^=0, A and ^7 hyperons have 8== 1 and S has
8^ 2. It is noteworthy that^ + andJT^ are not particle and anti
particle, as both have w~l. Each of them has an antibaryon, for
which n 1.
For K and Timesons, n = 0, since they are not baryons. This gives
8 =0 for remesons and S 1 for .fiT mesons. Unlike J7 =fc , TC* are related
as particle and antiparticle.
Then a selection rule is defined: the given interaction is strong
only when the resulting strangeness of all particles entering into the
reaction is conserved. For instance, every interaction of nucleons and
7rmesons only is strong (if it does not violate any other conservation
laws, except strangeness).
The simultaneous creation of A and K particles belongs to strong
interactions also, since one of them has /S = l, and the other S = 1.
But the spontaneous decay of the A particle into a nucleon and a
7rmeson is due to a weak interaction, because, here, the strangeness
is not conserved.
Transitions with AS = 2 are forbidden more strongly than with
= 1. That is why the S particle must decay first into a A orJT*
SeC. 38] THE RELATIVISTIC WAVE EQUATION 407
particle, which in turn decays into T nucleons and 7umesons. These
statements agree with the cascade nature of S decay.
There is no reason as yet to attribute definite values of T* and 8
to leptons, since they do not take part in strong interactions.
The transition to the nonrelativistic wave equation. It is instructive,
in comparing the relativistic wave equation with Schrodinger's equa
tion, to perform the limiting transition. We shall consider that the
energy of the electron is positive and that its velocity v is considerably
less than the velocity of light. Then <? differs from me 2 by a small
quantity^. If we take mc 2 ^ and mc 2 $ 2 to the lefthand side in the
first and second equations (38.11), the components ^ and ^ 2 are mul
tiplied by the quantity S we 2 , i.e., by y~. The wavefunction
components ^3 and ip 4 appear on the right, multiplied by cp x or by
cp y . It follows that, for a positive energy, the components i[/ 3 and ^ 4
are less than ^ and tp 2 in the ratio ^.The same follows from the second
two equations (38.11): the components ^ 3 and ^ appearjon the right
multiplied by ~2 me 2 , and ^ and ip 2 are on the left with a factor of
the order cp.
For negative energies the components ^ 3 and ip 4 are large, while
^ l and <p 2 are less in the ratio ~ .
Consequently, in the nonrelativistic limit Dirac's equation is re
duced to twocomponent form (as is required according to^Sec. 32) for
a description of particles with spin ^ .
Spin magnetic moment. We shall now show how spin magnetic mo
ment is obtained. It is first of all necessary to write down the Dirac
equation in an electromagnetic field. We know that for a transition
from the equation for a free particle to an equation for a particle in a
field it is necessary to replace the momentum p by p A, where A
is the vector potential, and the energy <? by S e<p (see Sec. 21).
Thus, the Dirac equation in the presence of an electromagnetic field
isof the form
^+pmc 2 ^. (38.18)
As was pointed out, the relations between the wavefunction com
ponents in the nonrelativistic limit without field appear as
[(P*  ipd ^2 + P* W >
408 QUANTUM MECHANICS [Part III
It is convenient to write down these relations, according to Sec. 32,
with the aid of the operators o, the definition of which involves the
factor [see (32.2), (32.8), (32.9)]:
(38 ' 20)
Here, the two small wavefunction components ^ 3 and ^ 4 are for short
denoted by <j/, and the two large components ^i and ^2 are called fy.
In addition, the momentum p is replaced by p A.
c
We shall also call ^mc* \$' , where $' is the energy value which
appears in the nonrelativistic theory. Then, after replacing p by
p A, we obtain from the first two equations of (38.11)
(38.21)
We can now eliminate <{/ from the equations, so that only the rela
tions for the large components ty remain.
Substituting $' in (38.21) from (38.20), we shall have
(f e<p)</ a,pAa,pA*. (38.22)
(e \
o,p A we must take into account
C ' ~
the commutation relations between the components of o and also
between p and A. Taking advantage of the fact that 6 a y a* ,
and calling p x A X P X ..., and analoguously for P y and P z ,
c
we shall first of all have
($' 69)^ g (p A 1 +  (S x a y P x P y \~ o y a x P y P x ) + . . . I ^ .
(38.23)
%
Further, we utilize the fact that 6*6 y = Oya*^^ [see (32.13)],
and also the commutation relations of the form
P X Py Py P X ~ ~ (p X Ay + A x fry fry A
c
(38.24)
e ^ ( ^ X _ ^^* \ e k
~~cT\~ay dx~) ~~~i^
[cf. (30.36)]. Substituting this in (38.23), we obtain an equation for
the components
Sec. 38] THE RE:LATIVISTIC WAVE EQUATION 409
(f  e<p) * = [~ (p  J A) 2  (5, #* + S,T, + S,ff,)] * .
Going over to vector notation, we arrive at the iionrelativistic
wave equation
Compared with the Hamiltonian operator for a spinless particle,
the Hamiltonian of an electron involves an additional term:
^o = ^(5H)s(i H). (38.26)
But since 6 is an additional mechanical moment, we see that the elec
tron has an additional magnetic moment
Ao=^S (38.27)
in accordance with what was affirmed in (32.17) (a here is a dimension
less operator). Spin differs from orbital angular momentum in that
its magnetic moment does not contain a factor 2 in the denominator.
Thus, the socalled spin magnetic anomaly follows naturally from the
Dirac equation.
The radiationfield correction to the magnetic moment. Equation
(38.27) is, of course, correct only in the nonrelativistic limit. But even
in this limit it is not completely exact. As was indicated in Sec. 27,
the state of an electromagnetic field in which there are no quanta
interacts with charged particles. Strictly speaking, insofar as there
is an interaction between the charges and the field, the state of each
separately cannot be defined with complete precision. It is therefore
not surprising that any state of a field is in some way perturbed by the
presence of charges, and any state of the charges is perturbed by the
field. As a result of this, the magnetic moment of an electron, as is
shown by the rather exact calculations of Schwinger and others, is
greater than one magneton by a very small quantity, whose relative
yj2
fraction is ^ r . This result is in complete agreement with experi
ment.
The magnetic moments of the proton and neutron do not at all
agree with the Dirac theory. For instance, on Dirac's theory, a neutral
particle (the neutron) should not have a magnetic moment at all.
In actual fact, the neutron possesses a magnetic moment directed
opposite to the spin.
This is usually explained by the strong interaction between the
nucleoii and the nuclear force field or, as it is sometimes called, the
meson field. There is a certain analogy here with the correction to
410 QUANTUM MECHANICS [Part III
the magnetic moment of the electron. This correction is small because
the interaction constant r is small. Nuclear interaction is very strong,
he
and so the result is a large "correction," if one can use that expression
for a quantity which, in the case of a proton, is twice as great as the
basic magnetic moment given by the Dirac theory.
At the present time we are unable to calculate the magnetic moment
of a nucleon, since no theory of nuclear forces exists.
Nevertheless, the nucleon is undoubtedly to some extend a Dirac
particle, as confirmed by the existence of the antiproton.
Energy eigenvalues of a hydrogen atom. In accordance with the
Dirac equation, the energy eigenvalues of a hydrogen atom, or of
any singleelectron atom, are calculated in the following way:
1, (38.28)
+
where n is the principal quantum number, j = l^,i.e., j is the total
electron angular momentum, a = ^ = y^ . If we regard ocZ as small
compared with unity, then the nonrelativistic formula (31.34) results.
It follows from formula (38.28) that the states 2^/ 2 and 2si/ a , with
the same n ~ 2 and j~^, have the same energy. In practice, these
states of the atom are somewhat split as a result of the interaction
of light quanta with the ground state of the field. The calculated
splitting agrees with experiment with considerable accuracy.
Exercises
1) Prove that a quantum cannot give rise to an electron positron pair
in free space in the absence of an additional external field.
The conservation laws in the absence of a field are written thus:
V 2 c* f c a p a 4 h <o = V 2 c 4 f c 2 pi , pf n ^pj.
c
Here, p is the electron momentum in a negative energy state, n is a unit
vector in the direction of the quantum momentum, p t is the electron momentum
in a positive energy state. Substituting Pj in the first equality and squaring
the left and righthand sides, it is easy to see that this equation is not satis
fied.
Another method of proof is based on simple reasoning. A transition to
another inertia! system can always make the energy of a quantum less than
2 me 2 . A quantum cannot give rise to a pair in such a system, simply because
it has insufficient energy. But what is impossible in one reference system
is impossible in all systems, because the possibility or impossibility of an
event does not depend upon the choice of the reference system.
Sec. 38] THE BELATTVISTIO WAVE EQUATION 411
The preceding argument no longer holds if pair production is considered
close to a nucleus. Here, the nucleus is at rest in one reference system and
in motion in another. Where the energy of the quantum is less than 2 me 2 ,
the moving nucleus will "help" it to give rise to a pair. Naturally, it is in no
way possible for a quantum to give rise to a pair if its energy in the rest system
of the nucleus is less than 2 me*.
2) Obtain the solution to the Dirac equation for a free electron.
Let us equate ^ to zero. Then the first equation of (38.11) is satisfied if
we take ^ 3 ^ Ac (p x ip y ), <J> 4 ~ Ac p z . The second equation of (38.11)
gives
Y2
me 2
The third equation of (38.11) reduces to the identity
(<T J me 2 ) ^ 3 = Ac (& f me 2 ) (p x ip y ) = c(p x ip y ) <j; a = Ac (p x  ip y ) (& \ me 2 ) .
The fourth equation of (38.11) also reduces to an identity. The number A
is determined from the normalization condition
A* [(* + we 2 ) 2 + c*p* x f
I me 2 )
The components ^ 3 and 4>4 are small compared with ^ 2 if v < c. Therefore
the solution corresponds to positive energy. Another solution with positive
energy is obtained if we take <; 2 = 0. Negative energy solutions are obtained
if we choose ^ 3 or ^4 = 0.
3) Show that from Dirac's equation there follows a chargeconservation
equation which is analogous to (24.16):
whore  *  2 =  <K  2 + K*  a + Ks I 2 +  h  2 
Write down equation (38.18) and its complex conjugate; multiply the
first by 4>* and the second by fy; subtract the second from the first and utilize
the Hermitian nature of the operators a and p.
4) Show that if ^ is a solution with positive energy <f , then p 2 <]> is a solution
with negative energy ^.
The equation for <J> is
+ me 2 (3 <
Whence
This proves that a negative energy solution cannot be avoided.
5) Prove that the operators
acting upon fourcomponent functions, are spin operators.
412 QUANTUM MECHANICS [Part III
We have
.2 A 2 .... h* . 2 h*
ax ~ j y a ^ oty oijjr  ay = r ,
.. ft ____ A 2 _ . & ~
CTA . ay =   a y a.z a. z a x =  a.y <x. x 1 jr <*z 9
so that the spin operators determined here possess all the required properties
(see See. 32). This can also be seen from the definition of d in terms of o and
p. Wo notice that the spin operators do not permute functions of the first
pair 4>i ^2 and of the second pair <> 3 , <^4 but make permutations only inside
each pair.
(>) Show that according to the Dirac equation only the sum of the orbital
and spin angular momenta and not each angular momentum separately satis
fies the angularmomentum conservation law.
Tho total angular momentum is defined as
yp x + gr a* a y .
We calculate the commutator with the Hamiltonian:
yp x  A_
he
y )  ? (a x p y a y p x \ p x ay  p y a vV )
The Hamiltonian is commutative also with the square of the total angular
momentum J 2   jj + 7y + J J . Tho integrals of motion are J 2 and J z , and
not M z , o 2 and JVf^, 3^ separately.
PART IV
STATISTICAL PHYSICS
lee. 39. The Equilibrium Distribution of Molecules in an Ideal Gas
The subject of statistical physics. The methods of quantum mechanics
et out in the third part make it possible, in principle, to describe
ny assembly of electrons, atoms, and molecules comprising a macro 
copic body.
In practice, however, even the problem of an atom with two elec
rons presents such great mathematical difficulties that nobody, so
ar, has solved it completely. It is all the more impossible not only
o solve but even to write down the wave equation for a macroscopic
>ody consisting, for example, of 10 23 atoms with their electrons.
Yet in large systems, we encounter certain general laws of motion
or which it is not necessary to know the wave function of the system
o describe them. Let us give one very simple example of such a law.
Ve shall suppose that there is only one molecule contained in a large,
ompletely empty vessel. If the motion of this molecule is not defined
Beforehand, the probability of finding it in any half of the vessel is
qual to 1/2. If there are two molecules in the same vessel, the prob
,bility of finding them in the same half of the vessel simultaneously
3 equal to \~\ = 1/4. The probability of finding all of a gas, consisting
>f N particles, in the same half of the vessel (if the vessel is filled with
;as) is (1/2) N , i.e., an unimaginably small number. On the average,
here will always be an approximately equal number of molecules in
>ach half of the vessel. The greater the number of molecules forming
he gas, the closer to unity will be the ratio of the number of molecules
n both halves of the vessel, no matter at what time they are
>bserved.
This approximate equality for the number of molecules in equal
volumes of the same vessel gives an almost obvious example of a sta
istical law applicable only to a large assembly of objects. In addition
o a spatial distribution, molecules possess a definite velocity distri
414 STATISTICAL PHYSICS [Part IV
bution, which, however, can in no way be uniform (if only because the
probability of an infinitely large velocity is equal to zero).
Statistical physics studies the laws governing the motion of large
assemblies of electrons, atoms, quanta, molecules, etc. The problem
of the velocity distribution of gas molecules is one of the simplest
that is solved by the methods of statistical physics.
Statistical physics introduces a series of new quantities, which can
not be defined in terms of singlebody dynamics or the dynamics of a
small number of bodies. An example of such a statistical quantity is
temperature, which turns out to be closely related to the mean energy
of a gas molecule. If a gas is confined only to one half of a vessel, and the
barrier dividing the vessel is then removed, the gas will itself uniformly
fill both halves. Similarly, if the velocity distribution of the molecules
is disturbed in some way, then, as a result of collisions between the
molecules there will be established a very definite statistical distri
bution, which, for constant external conditions, will be maintained
approximately for an indefinitely long time. This example involving
collisions shows that regularity in statistics arises not only because a
large assembly of objects is taken, but also because they interact.
The statistical law in quantum mechanics. Quantum mechanics also
describes statistical regularities, but relating to a separate object.
Here, the statistical regularity manifests itself in a very large number
of identical experiments with identical objects, and is in no way relat
ed to the interaction of these objects. For example, the electrons in a
diffraction experiment may pass through a crystal with arbitrary time
intervals and nevertheless give exactly the same statistical picture
for the blackening of a photographic plate as if they had passed through
the crystal simultaneously.
Regularities in alpha disintegration cannot be accounted for by the
fact that there are a very large number of nuclei: since there is practi
cally no interaction between nuclei inducing the process, the statisti
cal character predicted by quantum mechanics is only manifested
for a large number of identical objects ; it is by no means due to their
number. In this connection, a description of phenomena in quantum
mechanics involves the concept of probability phase, similar to the
concept of the phase of a light wave.
In principle, the wave equation can also be applied to systems con
sisting of a large number of particles. The solution of such an equation
represents a detailed quantummechanical description of the state
of the system. Let us suppose that as a result of the solution of the
wave equation we have obtained a certain spectrum of energy eigen
values of the system
* = *Q, *!,**, .>, ... (39.1)
in states with wave functions
Sec. 39] THE EQUILIBRIUM DISTRIBUTION OF MOLECULES 415
Then the wave function for any state, as was shown in Sec. 30, can
be represented in the form of a sum of ^functions of states with defi
nite energy values:
(39.2)
The square of the modulus
W n = \Cn\ 2 (39.3)
gives the probability (when the energy of a system in the state fy is
measured) that the result will be the nth value.
The expansion (39.2) makes it possible to determine not only ampli
tudes, but also relative probability phases corresponding to a detailed
quantummechanical description of the system.
The methods of statistical physics make it possible straightway
to determine approximately the quantities w n = c M  2 , i.e., the prob
abilities themselves omitting their phases. For this reason, the
wave function of the system cannot be determined from them, al
though it is possible to find the practically important mean values
of quantities that characterize macroscopic bodies (for example,
their mean energy).
In this section we shall consider how to calculate the probability
w n as applied to an ideal gas.
Ideal gases. An ideal gas is a system of particles whose interaction
can be neglected. The interaction resulting from collisions between
molecules is essential only when the statistical distribution w n is in
the process of being established. When this distribution becomes
established the effect of interaction is very slight.
As regards condensed (i.e., solid and liquid) bodies, the molecules
are all the time in vigorous interaction, so that the statistical distri
bution depends essentially upon the forces acting between the bodies.
But even in a gas the particles must not be regarded as absolutely
independent. For example, Pauli's principle imposes essential limi
tations on the possible quantum states of a gas. We shall take these
limitations into account when calculating probabilities.
The states of separate particles ol a gas. In order to distinguish
the states of separate particles from the state of the gas as a whole
we shall denote their energies by the letter e and the energy of the
whole gas by S. Thus, for example, if the gas is contained in a rec
tangular potential well (see Sec. 25), then the energy values for each
particle are calculated according to equation (25.19).
Let e take on the following series of values:
6 = 60,6!, e 2 ,...,e fe ,..., (39.4)
where there are n Q particles in the state with energy e and in general
there are nk particles with energy g* in the gas. Then the total
energy of the gas is
416 STATISTICAL PHYSICS [Part IV
(39.5)
By giving different combinations of numbers Uk, we will obtain
the total energy values forming the series (39.1).
We have repeatedly seen that the energy value Zk does not yet
define the particle states. For example, the energy of a hydrogen
atom depends only upon the principal quantum number n,* so
that the atom can have 2n 2 states for a given energy [see (33.1)].
This number, 2n 2 , is called the weight of the state with energy e,,.
But it is also possible to place the system under such conditions that
the energy value defines the energy in principle uniquely. We note,
first of all, that in all atoms except hydrogen the energy depends
not only on n, but on I also, i.e., on the azimuthal quantum number.
Further, account of the interaction between spin and orbit shows
that there is a dependence of the energy upon the total angular
momentum j and, finally, if the atom is placed in an external magnetic
field, the energy also depends upon the projection of angular momen
tum on the magnetic field. Thus, one energy value mutually corre
sponds uniquely to one state of the atom.
In a magnetic field all the 2n 2 states with the same principal
quantum number are split. We now consider how the states of a
gas in a closed vessel are split. We shall suppose that the vessel is
of the form of a box with incommensurable squares of the sides
#i> #!> #! Then, in accordance with equation (25.19), the energy
of the particles is proportional to the quantity  + + ~
where n v n& n a are positive integers. Any combination of these
integers gives one and only one number for the incommensurable values
af, a, a. Therefore, specification of the energy defines all three
integers n lt n& n$. If the particles possess an intrinsic angular mo
mentum, we can, so to speak, remove the degeneracy by placing
the gas in a magnetic field (an energy eigenvalue to which there cor
respond several states of a system is termed degenerate). We shall
first consider only completely removed degeneracy.
States of an ideally closed system. We shall now consider the energy
spectrum of a gas completely isolated from possible external influences
and consisting of absolutely noninteracting particles. For simplic
ity, we shall assume that one value of energy corresponds to each
state of the system as a whole, and, conversely, one state corresponds
to each energy value. This assumption is true if all the energy eigen
values for each particle are incommensurable numbers).* We shall
* Not to be confused with nfc!
* In a rectangular box the state e (n lf n 2 , n 3 ) has an energy which is com
mensurable with the energy of state e (2%, 2 n 2 , 2 n 3 ). Therefore, the energy
of all states can be incommensurable only in a box of more complex form than
rectangular.
Sec. 39] THE EQUILIBRIUM DISTRIBUTION OF MOLECULES 417
call these numbers fe. Then, if there are nk particles in the &th
state, the total energy of the gas is equal to $ =nkk . But, for
incommensurable fe, it is possible in principle to determine all
nk from this equation, provided < is precisely specified. It is clear,
however, that the energy of a gas consisting of a sufficiently large
number of particles must be specified with trully exceptional ac
curacy for it bo be possible to really find all nk from ff.
It is not a question of determining the state of an individual par
ticle from its energy <f , but of finding the state of the whole gas
from the sum of the energies of all of its particles. Every interval
of values d ff, oven very small (though not infinitely small), will
include very many eigenvalues <f . Each of them corresponds to its
own set of values nk, i.e., to a definite state of the system as a whole.
States of a nonidoally closed system. Energy is an exact integral
of motion only in an ideally closed system. The state of such a system
is maintained for an indefinitely long time, and the conservation
of the quantity $ provides for the constancy of all Uk. But nature
does not (and cannot) have ideally closed systems. Every system
interacts in some way with the surrounding medium. We will regard
this interaction as weak and will determine how it affects the be
haviour of the system.
Let us assume that the interaction with the medium does not
noticeably disturb the quantum levels of separate particles. Never
theless, every level efe ceases being a precise number and receives
a small, though finite, width A^. This is sufficient for the moaning
of the equation ffnkZk\>v change in a most radical way: in a
system consisting of a large number of particles, the equation con
taining approximate quantities k no longer defines the number Uk.
In other words, an interaction with the surrounding medium,
no matter how weak, makes impossible an accurate determination
of the state from the total energy <f .
Transitions between closeenergy states. In an ideally closed system,
transitions were forbidden for all states corresponding to an energy
interval d <? because the energy conservation law held strictly. For
weak interaction with the medium, all transitions that do not change
the total energy of the gas as a whole are possible to an accuracy
which is, in general, compatible with the determination of the energy
of a nonideally closed system.
Let us suppose that the interaction with the medium is so weak
that, for some small interval of time, it is possible, in principle, to
determine all the quantities nk and thus to give the total energy of
the gas $ =JT'wfefe. But over a large interval of time the state of
270060
418 STATISTICAL PHYSICS [Part IV
the gas can now vary within the limits of that interval of total energy
which is given by the inaccuracy in the energy of separate states
Ae*. Ah 1 transitions will occur that are compatible with the approx
imate equation = 57 nk (fe A efe) , Naturally , the state in which
k
all Ae* are of one sign is extremely improbable; this is why the
double symbol is written. We must find the state that is formed
as a result of all possible transitions in the interval dS .
The probabilities of direct and reverse transitions. A very important
relation exists between the probabilities for a direct and reverse tran
sition. Let us first of all consider this relation on the basis of equation
(34.29), which is obtained as a first approximation in perturbation
theory. Let there be two states A and B in the system, with wave
functions <{M and ^B The same value of energy corresponds to
these states, within the limits of inaccuracy d$ given by the inter
action of the system with the medium. In the interval dS, both
states may be regarded as belonging to a continuous spectrum.
Then, from (34.29), the probability of a transition from A to B in
unit time is equal to
where
27C
\ , and from B to A,
27T
(the weights of the states are denoted by QA, gu). But, if gA = QB,
then the probabilities for direct and reverse transitions, which we
shall call WAD and WBA> are equal because  J^AB\ 2 = \ <%?BA  2 . Natu
rally, the transition is only possible because the energies &A and
SB are not defined with complete accuracy, and a small interval
dff is given in which the energy spectrum is continuous. In an ideally
closed system we would have SA ^ <B.
The relationship we have found only holds to a first approximation
in the perturbation method. However, there is also an accurate
relationship that can be deduced from the general principles of
quantum mechanics. In accordance with the accurate relationship,
the probabilities for transitions from A to B and from B* to A*
are equal; here A* and B* differ from A and B in the signs of all
the linear and angularmomentum components.
The equal probability for states with the same energy. We have
seen that, due to interaction with the medium, transitions will occur,
in the system, between all kinds of states A, B, C, . . ., belonging
to the same energy interval d$. If we wait long enough the system
will pass equal intervals of time in the states A, B, C, . . .. This is
most easily proven indirectly, supposing first of all that the probabil
Sec. 39] THE EQUILIBRIUM DISTRIBUTION OF MOLECULES 419
ities for direct and reverse transitions are simply equal:
The refinement WAB = WB*A* does not make any essential change.
For simplicity we shall consider only two states such that WAB = WBA.
We at first assume that IA is greater than ts, so that the system
will more frequently change from A to B than from B to A. But
this cannot continue over an indefinitely long time, because if the
ratio IA: ts increases, the system will finally be constantly in A de
spite the fact that a transition is possible from A to B. Only the
equality tA=tB can hold for an indefinite time (on the average)
on account of the fact that direct and reverse transitions occur on
the average with equal frequency. The same argument shows that
if there are many states for which direct and reverse transitions
are equally probable, then over a sufficiently long period of time
the system will, on the average, spend the same time in each state.
We can suppose that fo= ^*, because the states A and A* differ
only in the signs of all linear and angular momenta (and also the
sign of the external magnetic field, which must also be changed so
that the magnetic energy of all particles is the same in states A
and A*). If we proceed from the natural assumption that IA= t^*,
then all the precinged argument can be extended to the case when
We have thus seen that the system spends the same time in all
states (with the same weight) that belong to the same total energy
interval d $.
The probability o! a separate state. We will call the limit of the
ratio tA/t, when t increases indefinitely, the probability of the state qA .
The equality of all IA implies that corresponding states are equally
probable. But this allows us to define the probability of each state
p
directly. Indeed, if there are P states, then T^^=l, because
^'
2^i U = t. But since the states are, as proven, equiprobable, we find
that qA = l/P. Similarly, the probability that a tossed coin will
fall heads is equal to 1/2, since the occurrence of heads or tails is
practically equiprobable.
Hence, the problem of finding probability is reduced to that of
combinatorial analysis. But in order to use this analysis we must
determine which states of the system can be regarded as physically
different. When computing the total number P we must take each
such state once.
Specification of gas states in statistics. If a gas consists of identical
particles, for example, electrons, helium atoms, etc., then its state
is precisely given if we know how many particles occur in each one
of the states. It is not meaningful to ask which particles occur in a
certain state, since identical particles cannot, in principle, be distin
27'
420 STATISTICAL PHYSICS [Part IV
guished from one another. If the spin of the particles is halfintegral
then Pauli's exclusion principle must hold and in each state there
will occur either one particle or none at all.
As an illustration for calculating the number of states of a system
as a whole, let us suppose that there are only two particles and that
each particle can have only two states a and b (ea=e&), each with
weight unity. In all, three different states of the system are con
ceivable :
1) both particles in state a; state b is unoccupied;
2) the same in state 6; a is unoccupied;
3) one particle in each state.
In view of the indistinguishability of the particles, state (3) must
be counted once because the interchange of identical particles between
states does not have meaning. If, in addition, the particles are subject
to the exclusion principle, then only the third state is possible.
Thus, if the exclusion principle is applicable to the particles then
the system can have only one state and, if it is not, then three states
are possible. Pauli exclusion greatly reduces the number of possible
states of a system. A system of two different particles, for example,
an electron and proton, would have four states because these particles
can obviously be distinguished.
Let us further consider the example of three particles occupying
three states. If Pauli exclusion is operative, then one, and only one,
state of the system as a whole is possible ; one particle occurs in each
separate state. If there is no exclusion, then the indistinguishable
particles can be distributed thus: one in each quantum state, or
two particles in one state and the third in one of the two remaining
states (this gives six states for the system as a whole), and all three
particles in any quantum state. Thus we have obtained 1 + 3 + 6 = 10
states for the system as a whole.
If these three particles differed, for example, if they were TU+,
TU, and 7rmesons of zero spin, then each of them could have any
one of three states independently of the others. Consequently, a
system of three such particles could, as a whole, have 3 3 = 27 states.
Later on we shall derive a general formula for calculating the mimber
of states.
Particles not subject to the Pauli exclusion principle. There is no
sense, for future deductions, in considering that each state of a par
ticle of given energy has unit weight. We shall denote the weight
of a state with energy e* by the letter gk. In other words, gk different
states of a particle have the same energy ei*. For every particle
these states are equally probable.
Let us assume that Uk particles have energy efe and are not subject
to Pauli exclusion. It is required to calculate the number of ways
that these nk particles can be distributed in gk states. We shall
Sec. 39] THE EQUILIBRIUM DISTRIBUTION OP MOLECULES 421
call this number Pg k n k . In accordance with what we have proved
above, the probability for each arrangement as a whole is
In order to calculate P gk n k we will, as is usual in combinatorial
analysis, call the state a "box" and the particle a "ball." The problem
is : how many ways are there of placing Uk balls in gk boxes without
numbering the balls, i.e., without desiring to know which ball lies
in which box. If the particles are not subject to the Pauli exclusion
principle then we must suppose that each box can accommodate
any number of balls.
Let us mix up all the boxes and all the balls so that we obtain
nk + gk objects. From these objects we take any box and put it
aside. The nk+gk 1 objects which remain are then randomly
taken from the common pile, irrespective of whether they are box
or ball, and placed in one row with this box from left to right. The
following series may be obtained:
bx, bl, bl, bx, bx, bl, bl, bl, bx, bl, bx, bl, bl, bx, bx, bx.
Since a box must appear on the left, the remaining objects can be
distributed among themselves (nk + gk 1)! ways.
We now throw each ball into the closest box on the left. In the
distribution we have used there will be two balls in the first box,
none in the second box, 'three in the third, one in the fourth, and
so on. There are (nk + gk 1)! distributions in all, but they are
not all distinguishable. Indeed, if we place the second ball in the
position of the first or any other one, nothing will change in the
series shown. There are nk\ permutations between the balls. In
exactly the same way the boxes can be interchanged with the boxes
because it does not matter in which order these boxes appear. Only
we must not touch the first box, because it always appears on the
left by convention. In all, there are (gk 1)! permutations of the
boxes. It follows that, of all the possible (nk + gk 1)! arrangements
in the series, only the following set of arrangements is different:
p __ (n k + gfc 1) I XOQAX

If, for example, nk=3, <7*=3 then P^ t = j= 10, which is
what we have already seen from direct computation.
Particles subject to Pauli exclusion. In the case of particles subject
to Pauli exclusion the calculation of P& k n k is still simpler. Indeed,
here we always have the inequality nk ^ gk, because not more than
one particle occurs in each state. Therefore, of the total number of
gk states nk are occupied.
The number of ways in which we can choose nk states is equal
to the combination of gk things nk at a time :
422 STATISTICAL PHYSICS [Part IV
(39 ' 7)
There are as many possible different states in the case of nk^gu,
and there is one particle or none at all in any of the gk states.
The most probable distribution of particles among states. The
numbers gk and Uk refer to a single definite energy. The total number
of states of a gas is equal to the product of the numbers P gk njt for
all the states separately:
So far we have only used combinatorial analysis. And besides it has
been shown that all separate states taken separately are equally
probable. The quantity P depends upon the distribution of particles
among the states. It can be seen that, in fact, a gas is always close
to a state where the distribution of separate particles among the
states corresponds to the maximum value of P possible for a given
total energy and for a given total number of particles.
We shall explain this statement by a simple example from gambling,
as is usually done in probability theory (most easily seen here is the
manifestation of largenumber laws in a game of chance). Let a coin
be tossed N times. The probability that it will fall heads once is
equal to 1/2. The probability that it will fall heads all N times is
equal to (1/2) N . The probability that it will fall N 1 times heads,
and once tails, is equal to (1/2) N ~ 1 x 1/2 x N, because this single
occasion can turn out to be anyone, from the first to the last, and
the probabilities for mutually exclusive events are additive. The
probability for a double tails is equal to (yj L_r_A 
The last factor shows how many ways two events can be chosen from
the total number N (the number of combinations of N two at a time).
In general, the probability that the coin will fall tails k times is
_ / j_w^ m fe
\2/ \2/ ~k\(
N\
\(Nk)\~'
The sum of all probabilities is, of course, unity:
because the sum of binomial coefficients is equal to 2 N .
Considering the series qk, we can see that qk increases up to the
middle of the sum, i.e., as far as k~N/'2, and then decreases sym
metrically with respect to the middle of the sum. Indeed, the kth
Sec. 39] THE EQUILIBRIUM DISTRIBUTION OF MOLECULES 423
term is obtained from the (k l)th term multiplied by
so that the terms increase as long as N/2>k.
Every separate series for tails appearing is in every way equally
probable with all other series. The probability for any given series
is equal to (1/2) N . But if we are not interested in the sequence of
heads and tails, but only in their total number, then the probabilities
will be equal to the numbers qk. For N^> 1, the function qk has a
very sharp maximum at k = N/2 and rapidly falls away on both
sides of N/2. If we call the total number of N tosses a "game," then
in the overwhelming majority of games, heads will be obtained ap
proximately N/2 times (if N is large). The probability maximum
will be sharper, the greater N is. We will not, here, refine this as
applied to the game of pitch and toss (see exercise 1), but will return
to the calculation of the number of states of a gas.
On the basis of the equal probability for the direct and reverse
processes between any pair of states, we have shown that any pre
viously defined distribution of particles among states has the same
probability of being established for a given total energy. In the
same way, every separate sequence of heads and tails in each separate
game is of equal probability. But, if we do not specify the states
of a gas by denoting which of the gk states with a given energy
are filled, and give only the total number of particles in a state with
energy fe, then we obtain a probability distribution with a maxi
mum similar to the probability distribution of games according to
the total number of occurrences of tails irrespective of their sequence.
The only difference is that in the example of pitch and toss the
probability depends upon one parameter &, and the probability
for the distribution of gas particles among states depends upon
all ttfe.
Our problem is to find this distribution for particles with integral
and half integral spins.
It is most convenient to look for the maximum of the logarithm
of P rather than P itself. In P is a monotonic function of the argument
and assumes a maximum value at the same time as the argument P.
Stirling's formula. In calculations we shall require logarithms
of factorials. For the factorials of large numbers, there is a con
venient approximate formula which we shall deduce here.
It is obvious that
n
lnn\ = In (1*2* 3*4 ... n) = JPln k .
The logarithms of large numbers vary rather slowly since the difference
ln(n + l) Inn is inversely proportional to n. Therefore, the sum
can be replaced by an integral:
424 STATISTICAL PHYSICS [Part IV
Inn! ~ J^ln& ^ J In kdk = nlnn n = rain . (39.9)
*i b
This is the wellknown mathematical formula of Stirling in somewhat
simplified form. It becomes more accurate the greater n is.
Additional maximum conditions. And so we must look for the values
of the numbers n* for which the quantity
8 = hi P = hi PI ^** (39.10)
k
is a maximum at a given total energy
& = n k zk (39.11)
k
And for a given total number of particles
N = nk  (39.12)
k
This kind of extremal condition is termed bound, because addi
tional conditions (39.11) and (39.12) are imposed upon it.
We shall first of all find nk for particles which are not subject
to Pauli exclusion, i.e., those with integral spin. To do this we must
substitute the expression for P from (39.6) in (39.10), take the dif
ferential dS with respect to all n^ and equate it to zero. We have
k k (39.13)
We substitute here the expression for factorials according to
Stirling's formula (39.9):
(39.14)
Since gr* is a large number, unity can naturally be neglected every
where. We must, of course, differentiate with respect to nk in formula
(39.14), because jr* is the given number of all states. Then
n k )  Inn*] = ^dn k \n^^ =0 .
* * (39.15)
It must not be concluded from this equation that the coefficients
of every dnk are equal to zero, because nk are dependent quantities.
Sec. 39] THE EQUILIBRIUM DISTRIBUTION OF MOLECULES 42f
The relationship between them is given by the two equations (39. IT
and (39.12) and, in differential form, are as follows:
Q > (39.16;
k
dN = dn k = 0. (39.17;
From these equations, we could eliminate any two of the numbers
drik, substitute them in (39.15), and afterwards regard the remaining
dnk as independent quantities. Then their coefficients may be regarded
as equal to zero.
The method of undetermined coefficients. The elimination of de
pendent quantities is most conveniently achieved by the method
of undetermined coefficients. This makes it possible to preserve
the symmetry between all Uk. Let us multiply equation (39.16;
by an indefinite coefficient which we denote by 1/0; the meaning
of this notation will be explained later. We multiply the second
equation (39.17) by a coefficient which we denote (j,/0, so that we
have introduced, as is required from the number of supplementary
conditions, two quantities, and pi. After this we combine all three
equations (39.15)(39.17) and regard all dnk as independent, and
6, and (JL as unknown values which should be determined from equations
(39.11) and (39.12). The maximum condition is now written as
dS  **. + **L = o. (39.18;
We look for the extremum of one quantity S  + ^ > anc
then choose and \L so that the energy and number of particles
equal the given values. But if the extremum is determined foi
one function without conditions, then all its arguments become
mutually independent, and we are entitled to equate any differential
to zero regardless of the other differentials.
Equation (39.18), written in terms of dnk, has the following form:
BoseEinstein distribution. Let us now put all the differentials
except dnk equal to zero. According to what we have just said this
is justified. Then, for equation (39.19) to hold, we must put the
coefficient of dnk equal to zero:
AJt^o. (39.20)
x '
426 STATISTICAL PHYSICS [Part TV
Naturally, this equation holds for all 1c. Solving it with respect
to rife, we arrive at the required most probable distribution of the
number of particles according to state:
. e *r^ ' (39.21)
e 1
This formula is called the BoseEinstein distribution. As to particles
for which this distribution is applicable, they are said to obey Bose
Einstein statistics or, for short, Bose statistics. They have either
integral or zero spin. The unknown quantities and [JL, i.e., the para
meters in the distribution, are given by equations as functions of
N and S\
<** ' (39.22)
k 6 _!
y __ 0fc =AT
^ c fe* ' (39.23)
" .""I
so that the problem posed of finding the most probable values of
Tife is, in principle, solved.
FermiDirac distribution. We shall now find the quantities rik
for the case when the particles are subject to Pauli exclusion. In
accordance with (39.7) and Stirling's formula we have for the
quantity 8:
fl = In PI 77^77 =
1 n! w!
09.24)
k
Differentiating $ and substituting into equation (39.18), we obtain
k
whence, by the same method, we arrive at the extremum condition:
and the required distribution appears thus:
fc = ^gr ( 39  26 )
e * +1
Sec. 39] TBOB EQUILIBRIUM DISTRIBUTION OF MOLECULES 427
Here, nk < gk as is the case for particles subject to Pauli exclusion.
For such particles, formula (29.36) is called a FermiDirac distri
bution. The parameters 9 and fji are determined analogously to (39.22)
and (39.23):
= <f , (39.27)
(39.28)
Concerning the parameters 6 and p. The parameter is an essentially
positive quantity, because otherwise it would be impossible to satisfy
equations (39.22), (39.23) and (39.27), (39.28). Indeed, there is no
upper limit to the energy spectrum of gas particles. For an infinitely
1*.
large e and < 0, we would obtain e e = , so that, by itself, a Bose
distribution would lead to the absurd result nk < 0. In (39.23),
on the left, we would have the negative infinite quantity^ yk,
which can in no way equal N. Similarly, a Fermi distribution would
lead to infinite positive quantities on the lefthand sides of (39.27)
and (39.28); and this is impossible for finite N and 8 on the right.
Therefore,
0>0. (39.29)
In the following section it will be shown that the quantity is pro
portional to the absolute temperature of the gas.
The quantity \i is very important in the theory of chemical and
phase equilibria. These applications will be considered later (see
the end of Sec. 46 and the succeeding sections of the book).
The weight of a state. Here we give a few more formulae for the
weight of a state of an ideal gas particle. The weight of a state with
energy between e and s + de is given by the formula (25.25), whose
lefthand side we shall denote now by dg (e). In addition we assume
that the particles have an eigenmoment j, so that we must take
into account the number of possible projections of j, equal to 2j+l:
. (39.30)
For electrons ? = l/2, so that 2; + 1=2.
For light quanta we must use formula (25.24), replacing K in
it by co/c and multiplying by two, according to the number of possible
polarizations of the quantum:
(39.31)
428 STATISTICAL PHYSICS [Part IV"
It is also useful to know the weight of a state whose linear momen
tum is between p x and p x + dp x , p Y and p y + dp y , p* and p z + dp z .
It is determined in accordance with (25.23), also with account taken
of the factor 2; + l. Thus, for electrons, we obtain
. (39.32)
Exercises
1) Write clown the formula for the probability that heads are obtained k
times for largo N, whore k is close to the maximum (jk>
Tho general formula for probability is of the form:
Nl _ N
'# (Nk)lkl '
Wo shall consider the numbers N and k us largo. It is more, convenient hero
to use, Stirling's formula in a somewhat more exact form than (39.9), namely:
Nln N ~ + ~ln i 2KN.
c 2
N N
Wo put k ~ _ f *^ N k =  .* , where x is a quantity small com
^ LJ
N
pared with . Then, in the correction terms of Stirling's formula 1/2 In '2xk
and 1/2 In 2 K (N  ), the quantity jr win bo neglected. Wo expand the, denom
inator in a series up to x z :
11, 1 / N , \ , ^ 1 ^ , 1 ^ V , ^ , l 1 O ^
Infc! In I g^ + ^l! =  ~ m ^<T + J? In   + ^ 'I y ln 2^^.
Tho corrootion terms aro
Substituting this in the expression for qk and taking antilogarithms we
arrive at the required formula:
_ __ 2.x*
]/ 2 N
* M ra e
g has a maximum at x and dies away on both sides, q is reduced e times
1 /~y~
in the interval av = I  characterizing the sharpness of the maximum.
Compared with the whole interval of variation x, the interval x e comprises
Xe ir"2~
= I/   . p or example, for N ~ 1,000, the maximum is approximately
equal to 1/40. The ratio ' is about 2% so that, basically, the heads fall between
475 to 525 times. Tho probability that heads (or tails) will fall 400 times out
' 2 '
_
of a thousand is equal to 1/40e lt000 l/40e~ 20 . In other words, it is
SeC. 39] THE EQUILIBRIUM DISTRIBUTION OF MOLECULES 429
e +2 , i.e., several hundreds of millions of times less than the probability for
heads appearing five hundred times.
2) Verify that the probability q has been normalized, i.e., that \q (x) dx = 1 .
Since the probability decreases vory rapidly with increase in the absolute
value of x, the integration can bo extended from eo to oo without notice
able error. Then
Wo shall now show that the integral appearing hero is indeed equal to \/ re
We shall call it /:
Squaring, we get
 5 'd5 jV^  J
The integration spreads over tho w'lolo ^/jplano. Let us go over to polar coor
dinates :
= p cos 9 , 7) = p sin 9 .
Instead of d^dri we must put p</p dy, as is usual for polar coordinates.
Then,
27T
= f p6 P I ,/p f
or
so that
00
3) Find the mean square deviation, for the occurrence of heads, from the
most probable number, i.e., x 2 .
We have
00 00 ^ 2
2 _ I ~,'l n {~\ f?r / ( vIp N~ <7'r 
X ~~~ I *' (J {*'/ U"*' I/ ^y" I *' t> z * Ctt/ 
In order to calculate the integral we make use of the result of exorcise 2,
writing 2 a in the exponent instead of ^ 2 .
(vrdsy* .
J Va
430 STATISTICAL, PHYSICS [Part IV
Wo differentiate both sideH of this equation with respect to a:
2a''
oo
Putting a 1 we arrive at the required formula:
Wo notice, incidentally, that
and in general
For x a , we obtain
Expressing JV in terms of x' 2 , wo can write down the probabilitydistribution
law :
__*L
q(x)dx=   ~ ^
V 2
Thus, the width of the distribution is very simply related to the mean
square deviation of the quantity from its most probable value*:
. _ . A/ O ^2
j. g v i* *L .
Of courso, this relationship holds only for an exponential distribution of the
typo obtained in exorcise 1.
See. 40. Boltzmaim Statistics
(Traiislutional Motion ot a Molecule. Gas in an External Field)
Boltzmann distribution. Long before the Bose and Fermi quantum
distribution formulae (39.21) and (39.20) had been obtained, Boltz
maim derive J a elassical energy distribution law for the molecules
of an ideal gas. This law is obtained from both quantum laws by
means of a limiting process. We shall perform this transition purely
formally at first, and. then decide which real conditions it corresponds to.
* For .r = a\>, the probability q (.r) decreases e times compared with q (0);
it is for this reason that .i> characterizes tho width of the distribution.
OC. 40] BOLTZMANN STATISTICS 431
Let e be measured from zero, and let the ratio (ji/0 be large in absolute
alue and negative. Then
\ considerably greater thftii unity for all e. Here, unity in the deiiom
lator of both distribution formulae can be neglected as compared
'ith the exponent, and both the Bose and Fermi formulae take
n the same limiting form
Wfc = <7*' . (40.1)
'his is the Boltzmanii distribution. Let us now determine the constant
from the normalization condition for the distribution :
JTwfc=iV. (40.2)
k
Let us suppose that the gas molecules possess some internal degrees
f freedom (in addition to the external transport degrees of freedom)
lat may be related to electron excitation, the vibration of nuclei
dth respect to each other, and the rotation of the molecule in space.
'ho energy of all these degrees of freedom is quantized. Without
efiiiing it more exactly for the time being, we can write the total
tiergy of a molecule e in the form of a sum of the energies of trans
itional and internal motion:
Accordingly, the weight of a state of given energy is also represented
3 the product of two weights: one relates to translation al motion
tid is given by the formula (39.32), while the other we denote simply
y </(/) (we also agree to include in it the factor 2^ + 1):
8  <!<>
herefore, formula (40.2) can be written thus:
(40.5)
2
I I
Expanding the translational motion energy into V* + PY ~*~ P* ^
o see that the momentum integral is represented as the product
f three integrals of the form
2mO
dp x
432 STATISTICAL PHYSICS [Part IV
These integrals are easily calculated from the second formula
of exercise 3, Sec. 31). Each of them is equal to \/27rw6, so that con
dition (40.5) reduces to the form
l*Z<l&e" r  (40.6)
t
If the gas is monatomic then the quantities (1) refer to electron
excitations. Therefore, if sW > 0, then, actually, only the zero term
appears in the summation over the states*. But since the energy
is measured from ^ as from zero, the whole summation, actually,
reduces only to the; zero term #(<>). It is of the order of unity. For
example, when the ground state* has angular momentum 1/2, g^ )=^2.
We then obtain the condition for the applicability of Boltzmaim
statistics in the form
For the inequality (40.7) to be satisfied, it is sufficient to satisfy
one of two conditions:
1) the density of the gas is very small, i.e., the volume occupied
by the gas at a given temperature is large;
2) the temperature for a given volume V is very high.
In the case when the gas is nob monatomic, these conditions are
__.i (l !
quantitatively changed somewhat because (J(i) e is also some
i
function of 0. But qualitatively, the conditions of applicability of
Bolt'/mann statistics still hold.
Classical and quantum statistics. We have seen that for small
densities or high temperatures the quantum distribution laws for
a gas pass into the classical Boltzmaim law. From now on we shall
agree to call the Bose and Fermi statistics quant am statistics and
the Boltzmann statistics, classical, regardless of wheather the energy
spectrum is discrete or continuous. Those statistics will be termed
quantum for which the indistinguishability of separate particles is
taken into account. In other words, a quantum definition of the state
of a system lies at the basis of quantum statistics: the number of
particles in all quantum states must be given. The classical definition
of the state of a system indicates which particles are found in the
given states. The Boltzmann formula (40.1) can be obtained from
this classical definition.
Maxwell distribution. In this section we will not be concerned with
the statistics of the internal motion of molecules, and will consider
* Tho relation between and temperature is given by formula (40.25).
Sec. 40]
BOLTZMANN STATISTICS
433
only their traiislational motion. In accordance with (40.3), the energy
of the traiislational motion of molecules is separable from their
internal energy. Therefore, the Boltzmann distribution breaks up
into the product of two factors. We are not interested in the first
of the two factors, but the second, relating to traiislational motion,
is of the form
p %
f>~ ^tnQ
The weight of a state relating to a given absolute value p is obtained
by changing to polar coordinates in formula (39.32):
dg(p) V ff* (40.8)
[cf. (25.24)].
Thus, the distribution according to the kinetic energies of transla
tional motion is written in the form:
dn(p) =
(40.0)
It is applicable both to monatomic and polyatomic gases if m is the
mass of a molecule as a whole.
The constant A is found from the normalization condition
00
f  p '
I p*e * m *dp = N.
(40.10)
The value of the integral was found in exercise 3, Sec. 30. From this
vve obtain
. (40.11)
, !n place of the momentum distribu
*.i n.i of molecules, it is sometimes useful
:, ',ave their velocity distribution. For
i,ti ; ~ it is sufficient to substitute p = mv
i?e distribution (40.9):
(40.12)
Fig. 46
distribution had already been deduced by Maxwell, before
, and for this reason it is called the Maxwell distribution.
).?< Tig. 46 we have plotted the ratio ^^ on the ordinate. For
w O/D
, this quantity is close to zero because of the factor v 2 in the
 0360
434 STATISTICAL PHYSICS [Part I\
equation for the weight of a state ; after the zero point it reaches a
maximum and exponentially decreases to zero again for large velocities,
We thus sec that a gas contains molecules with every possible velocity
value.
The velocities of gas molecules. The greatest number of molecules
have a velocity corresponding to the maximum of the distribution
curve shown in Fig. 46. This maximum is determined from equatior
(40.12). The corresponding velocity is termed the most probable;
it is
(40.13;
We find the mean velocity by calculating the integral (we omit
the factor N, because the mean value of velocity relates to a single
molecule) :
^
The mean square velocity is also interesting
: =J^. (40.15
m v
(the result of exercise 3, Sec. 39 is used in the derivation).
The ratio Vv*~:v: v m . p . = VT : I/ : VT.
r TT
The mean energy of a single molecule is equal to
= ~ = G, (40.16;
and the mean energy for the whole gas is N times greater: /
3 f
<? = Nz = Y^ e  (40.F7
This result relates to the energy of translational motion of the mole
cules. Numerical evaluations of velocity will be performed below.
The relationship between energy density and pressure. We fshal
now derive a very important relationship between the density o:
kinetic energy of a gas and its pressure. This relationship holds for an}
statistics and depends only upon the form of the expression for eiaergj
in terms of momentum.
The pressure of a gas is defined as the force with which the gas act?
upon unit area perpendicular to its direction. This force is eqi al tc
the normal (to the surface) component of momentum transmitted by the
Sec. 40] BOLTZMANN STATISTICS 435
gas molecules in unit time. Let the direction of the normal to the sur
face coincide with the #axis. We first choose those molecules which
have a velocity component along the #axis equal to v x . They will
reach the surface of a volume in unit time if they initially were situated
in a layer of width v x . Let us cut out a cylinder from this layer with
base of unit area and height equal to v x . The volume of this cylinder
is v x . If dn (v x ) is now the number of molecules whose velocity compo
nent normal to the surface is v x , then the density of these molecules
is ~. There are v x n  such molecules in a cylinder of volume v x .
Each of them, upon elastically colliding with a wall, will reverse its
normal velocity component, and the wall will receive a momentum
mv x ( mv x ) = 2mv x . (40.18)
Thus, all the gas molecules having a velocity v x , transfer to the wall
in unit time a momentum
(40.19)
In order to obtain the gas pressure on the wall we must integrate
(40.19) over all v x from to oo, and not from oo to oo, because
molecules moving away from the wall will not strike it. Thus, the pres
sure of the gas on the wall is
(40.20)
oo
On the other hand, the mean kinetic energy of the gas is
CO OO ' OO
~f = f J vldn () + f J v$dn (,) + f J t d n (v,) =
), (40.21)
because the mean values of the squares of all the velocity components
are identical.
Comparing now (40.20) and (40.21) we find that the gas pressure is
equal to two thirds of the density of its kinetic, energy :
? = !' (40.22)
This result was published by D. Bernoulli, as early as 1738, a century
and a half before statistical physics began to develop as an independ
ent science.
28*
436 STATISTICAL, PHYSICS [Part IV
Only two assumptions have been used in the derivation of (40.22):
identical values of the three velocity projections are equiprobable
and the kinetic energy is equal to ~ . The concrete form of the
distribution function is not essential.
The Clapeyron equation. If a gas is subject to Boltzmann statistics,
then, in accordance with (40.17), the mean kinetic energy <?is equal
to 2" Substituting this in (40.22) we obtain
pV = NQ. (40.23)
But from the definition of absolute temperature
pV = RT. (40.24)
From this we obtain the relationship between ''statistical" temper
ature 0, measured in ergs, and the temperature T, measured in
degrees Centigrade:
fi _ J?_ T '* x 10? T  1 38 y 10 16 T (40 25}
N 6.024 x 10 23 L ' m X 1U  1 ' {w.AO)
7?
The ratio k=^is called Boltzmann's constant. It is equal to
1.38 x!0~ 16 . The temperature can also be measured in electronvolts,
one electronvolt being equal to 1.59 x 10~ 12 erg. Translating ergs into
degrees with the aid of Boltzmann's constant, we find that 1 ev~
=11,600.
As is known, the specific heat of an ideal inonatomic gas is equal
O
to '  jR, thus corresponding to an energy '  RT. Replacing RT by
NQ, we find ~g' = ~NQ in agreement with (40.17).
JL
The relationship (40.25) allows us to calculate the mean velocity
of gas molecules without using the Avogadro number N. Indeed,
where M is the molecular weight of the gas. For example, the mean
velocity of hydrogen molecules at a temperature of 300 K is
This value is comparable with the exit velocity of a gas into a
vacuum or with the velocity of sound [see (47.30)].
The thermonuclear reaction. When nuclei collide reactions are pos
sible between them that proceed with the release of energy. For exam
ple, in a deuterondeuteron collision one of two reactions can occur
(besides elastic scattering):
Sec. 40] BOLTZMANN STATISTICS 437
Here //* is tritium and n l a neutron. Another example is
In order that charged nuclei may be able effectively to collide, they
must overcome the potential barrier of Coulomb repulsion, which was
considered in Sec. 28. The dependence upon energy for the probability
of passing through the potential barrier is basicaUy determined by the
barrier factor [see the first term on the right in (28.12)]:
_ 2 TC Zj Z t e*
e ''"H . (40.26)
Here, Z : e and Z 2 e are the charges of the colliding nuclei and v\\ is the
relative velocity along their joining line [recall that (28.12) refers to
onedimensional motion].
The reaction can be produced by accelerating the particles in a dis
charge tube. But charged particles, striking a substance, mainly
spend their energy on ioiiization and excitation of the atoms. And
since, according to (40.26), the probability of the reaction at small
energy is vanishingly small, the majority of incident particles do not
cause a reaction. Of the total number of particles it turns out that
10~ 5 10~ 6 are effective. Therefore, the energy yield of the reaction
is considerably less than the total energy spent in accelerating the
beam of particles.
The situation is different if the substance used for the reaction is
at a very high temperature, of the order of 10 7 degrees. At this tem
perature, the nuclei of the heated substance already react at a suffi
ciently high rate, and transmission of energy to electrons does not
occur because their mean energy is the same as that of the nuclei.
Let us calculate the rate of a nuclear reaction occurring under such
conditions. It is termed thermonuclear.
Let the effective crosssection for the reaction between nuclei with
relative velocity v\\ be <r (VH). We assume that different nuclei react:
we shall call them 1 and 2. Let us construct on each nucleus 2 a cylinder
with base area a (v\\) and height numerically equal to v\\. Then, by
definition of a (v\\), all those nuclei 1 which occur in the volume of
these cylinders and which have velocity v\\ relative to nuclei 2 will
be involved in the reaction in unit time.
The number of such events in unit volume and unit time is equal
to the product of
(40.27)
where n x and n 2 are the numbers of nuclei 1 and 2 in unit volume, and
dq (v\\) is the probability that the relative velocity is equal to v\\.
438 STATISTICAL PHYSICS [Part IV
Indeed, a cylinder of volume v\\ G (v\\) can be constructed on each
nucleus 2, and there will be n a (v\\) v\\ nuclei 1 in each cylinder. The
velocity distribution of the nuclei is taken into account by multiplying
by dq (VH). If 1 and 2 are identical nuclei, then expression (40.27)
must be halved so that each reaction is not taken into account twice.
We indicate this by the factor (2) in the denominator of expression
(40.28).
Let us now determine the probability factor dq (v\\). The absolute
velocity distribution is given by the product of two Maxwellian fac
tors of the form
r^v* 1 ^.
In the exponent of this expression is the sum of the energies of both
nuclei. In accordance with formula (3.17), it can be split into the kine
tic energy of the motion of the centre of mass of the nuclei and the
kinetic energy of their relative motion. Hence, in the product a factor
is separated that gives the relativevelocity distribution:
mv l
~
20 26 26
where m == i^ [see (3.20)] , v 2 === v\ + vl .
m l 4 m 2 L v /J n a.
Normalizing the distribution over v\\ to unity and passing to the re
duced mass m, we obtain an expression for the probability that the value
of relative velocity along the line joining the nuclei will be v\\.
20
The barrier factor (40.26) depends upon v,,.
Thus, the overall rate of a thermonuclear reaction is
% n 2 f
= & J
events
Taking into account the barrier factor, we write the dependence of
effective crosssection upon the rate as
""
The factor a here depends considerably less upon the rate than the
exponential function.
Sec. 40] BOLTZMANN STATISTICS 439
The integral in (40.28) now reduces to the form:
I d (v { \) v\\ e hv]{ 20 dv l{ . (40.29)
o
It can be calculated, to a good approximation in the case when the
temperature is so low, that the greater part of the reaction proceeds
at the "tail" of the Maxwellian distribution at a rate greater than the
mean. Let us show how this calculation is done.
We denote the argument of the exponential in the integrand thus :
' 2 e 2 , w*V __ , *> M t
._, S _.
We find the minimum of the function / (v\\) from the condition
& =  4 + 6 = 0; V j = j/T . (40.30)
av \\ v\\ 1 b
We shall see that the basic contribution to the integral is given by
values of v\\ close to v\. Near the minimum, / (v\\) can be represented
in the form
0?,) 2 , (40.31)
and the integral (40.29) is written as
"" "" dv,,. (40.32)
The minimum of / (v\\) corresponds to the rate v*\ at which the
greatest number of reactions occur. The ratio of the rate vjj to the
mean relative rate  v\\ \ is, according to (40.30),
V I/ w a / 8 / 7  /Jl
J=T = \~2 VaVb =\2
bill v * x
since
o
o
v\\
We shall call the temperature low if the ratio is several times
I ll I
greater than unity. Then the maximum of the integrand of (40.32)
440 STATISTICAL, PHYSICS [Part IV
is very sharp at the point v\\ = v\, because it decreases e, times when
v\\ deviates from i?J by an amount \/^r , which is considerably less
than vjj.
It was therefore justifiable to terminate the expansion in (40.31)
with the second term. In addition, the quantities <T O (v\\) and v\\
can be taken outside the integral sign when v\\ = v*\. The error in
both approximations is of the order J ^ .The integration can be taken
F1
from oo to oo because the integrand rapidly decreases as v\\ re
cedes from v\\ 9 so that the error is exponentially small.
Thus,
. ,
Ja(t;,,)t;,,c
o
> , O v o
r "~~2" (t/ ii~~ l 'ii )
J e 2
(40.34)
Substituting the values a and 6 and using (40.28), we find the
expression for the rate of a thermonuclear reaction
(40.35)
The exponential factor depends very strongly upon the temperature.
For example, for a reaction in deuterium, this factor changes 3,600
times when the temperature is increased from 100 to 200 ev.
Thermonuclear reactions are the source of stellar energy and
for this reason play as important a part in our life as chemical
reactions !
Ideal gas in an external field. We shall now consider an ideal gas
acted upon by an external field with potential U. The potential ener
gy can depend both upon the position of the molecules in space as well
as their orientation (if the gas is not mon atomic).
The total energy of a molecule is
e =  + e(0 + u ( 4  36 )
If U depends upon the position of a molecule in space, i.e.,
U=U (#, y, z), then we must pass from a finite volume F, in the weight
factor (40.4), to an infinitely small volume dV dxdy dz. Then part of
the distribution function that depends upon coordinates x, y, z can be
Sec. 40] BOLTZMANN STATISTICS 441
separated, and a formula is obtained defining the dependence of gas
density upon coordinates:
dn(x 9 y,z) = n Q e dxdydz. (40.37)
Here, the potential energy calibration is U (0, 0, 0) = 0, and the gas
density at this point is equal to n . For example, in a gravitational
field, U = mgz, so that
dn(z) = n e dz . (40.38)
It should be noted that in the earth's atmosphere the "barometric"
formula (40.38) is rather more applicable qualitatively because air
temperature is not constant with height.
In addition, the "barometric" formula indicates that the composition
of the air must vary with height as a result of the different molecular
weights of nitrogen, oxygen, and other gases. Actually, the air com
position is almost uniform vertically because of vigourous mixing
processes.
The nonequilibrium state of planetary atmospheres. In place of the
approximate expression for the potential energy in a gravitational
field, let us substitute its exact expression (3.4). Let us first of all
express the constant a in formula (3.4) in terms of more convenient
quantities. The force of gravity at the earth's surface is mg and,
from the general gravitational law, it is equal to c \ , where r is the
m r
radius of the earth. From thisa = ragrr,sotliat J7= mffr . Therefore,
the gas density must vary with height according to the law
n^n^e 2r . (40.39)
This quantity remains finite even at an infinite distance from the
earth, and since the exponent is equal to unity at infinity we have
called the proportionality factor n^ .
Near the earth, where r = r Q , the density is greater than n^ by as
many times as the quantity
is greater than unity.
The radius of the earth r & 6.4 x 10 8 cm, g & 10 3 cm/sec 2 . From
this we obtain for oxygen
82 10* 6.4.10*
'
RT = 8.3 10' 300
442 STATISTICAL PHYSICS [Part IV
In actual fact the density of the earth's atmosphere at infinity is
equaJ to zero. Therefore, it follows from formula (40.39) that the
atmosphere cannot arrive at the most probable state when in the
earth's gravitational field, and is gradually dispersed into space.
The most probable state of a gas is called statistical equilibrium
(see Sec. 45). The equilibrium density of the atmosphere at infinity
is e 800 times less than at the earth's surface. Therefore the present
state of the atmosphere is very close to equilibrium. For the nioon,
equilibrium has been reached: its atmosphere has completely
escaped !
A kinetic interpretation of the dispersion of planetary atmospheres.
It is easy to understand the reason for the recession of gases to infinity.
Any particle whose velocity exceeds 11.5 km/sec is capable of over
coming the earth's attraction: its motion is infinite. In accordance
with the Maxwell distribution (40.12) a gas will always have molecules
with every possible velocity. In literal notation, the velocity of mole
cules capable of going to infinity is defined by the equation
(40.40)
Taking v 2 from this equation and substituting into the Maxwell
_ WtfTO^
distribution, we once again obtain the exponential e Q for the
fraction of molecules capable of leaving the atmosphere. It is easy
to estimate the number of such molecules in the atmosphere at any
instant of time. The earth's surface is 5xl0 18 cm 2 . There is about
1,030 gm of air above every square centimetre, i.e., about 35 mo]es.
Hence, the total number of molecules in the atmosphere is 5 x 10 18 x
X 35 x 6 x 10 23 = 10 44 , and the fraction of molecules of velocity greater
than 11.5 km/sec is e~ 800 ~10~ 344 . Therefore the mean number of mole
cules capable of leaving the earth at each instant is only 10~ 300 . Of
course, those molecules close to the earth's surface will not be able to
"carry" their energy to the upper layers of the atmosphere because
of collisions with other molecules.
The dielectric constant of a gas. We shall now consider a gas whose
molecules have a constant dipole moment in a constant and uniform
electric field. Those molecules can have characteristic dipole moments
for which there is some preferred direction: NO, CO, H 2 O (along the
altitude of the triangle passing through 0), NH 3 (along the axis of
symmetry of a threesided pyramid). The more symmetrical mole
cules do not possess moments: H 2 , O 2 > CH 3 (tetrahedron), CO 2 (this
proves that the C0 2 molecule has the form of a rod with the C atom
in the centre).
Rotational motion is quantized. In the next section it will be shown
that for all gases except hydrogen, at a temperature of several tens of
degrees (from absolute zero), the states with large quantum numbers
Sec. 40] BOLTZMANN STATISTICS 443
are already excited. In these states the motion may be rega ded as
classical. Then the total rotational energy of a molecule simply breaks
down into the kinetic energy of rotation (see Sec. 9) and potential
energy, which depends upon the orientation of the dipole moment
relative to the external electric field:
[see (14.28)]. In classical motion, the potential and kinetic energies
may be regarded as quantities instead of operators. Therefore, in the
Boltzmann distribution the factor that depends only upon potential
energy is split off:
d E cos &
dn(b) =Ae sin&d&. (40.41)
Here, sin ft d$ is proportional to the element of solid angle in which the
vector d lies [cf. (6.15)].
Let us now determine the electric polarization of a gas in an external
field. For this we must calculate the mean projection of the dipole
moment upon the electric field, i.e.,
'/
e
n dEcosQ
/.
(40.42)
It turns out that it is sufficient merely to find an expression for
the integral in the denominator, because equation (40.42) can be
rewritten thus:
(40.43)
Indeed, differentiating the integral with respect to the parameter E,
we revert to (40.42). The integral can be calculated using the fact that
sin & d& = d (cos 9) :
/
. * / At\ A A\
= sr7Sinh7r. (40.44)
The integral in formula (40.43) is called a statistical integral. For
quantized energy levels, it is replaced in the general case by a statisti
cal sum. The expression for the summation and integral will be met
with many times again. It is very convenient in calculating mean
values.
444 STATISTICAL PHYSICS [Part IV
Substituting (40.44) in (40.43), we obtain an expression for the mean
projection of the dipole moment on the electric field
(//' /I A \
coth^). (40.45)
This expression was obtained by Langevin. Let us investigate the
righthand side in two limiting cases: E<^ 7 (weak field) and E^> j
(strong field).
If the field is weak then we can use the expansion of coth x in terms
of x:
1 , x
X
whence
SIT (4 ' 46)
The polarization of the gas is
P = Nd cos & =  ^ , (40.47)
and the dielectric constant is calculated from the definitions of induc
tion (16.23) and (16.29):
D = E + 47r P = #l +  = *E. (40.48)
In a strong field coth ^ tends to unity and ^y tends to zero.
Therefore, cos tends to unity. This means that all the dipoles are
orientated along the external field and saturation sets in. Then
D = E + 4nNd. We notice that, for J 7 = 300K, this case would
correspond to a field E > 10 7 v/cm, which is considerably greater than
the breakdown potential.
Paramagnetism of gases. We shall now find how the magnetic per
meability x is calculated. Here we must take into account the fact that
the magnetic moment is related to the mechanical moment of electrons,
and the latter is a quantized quantity, i.e., it takes on a discrete
series of values. Usually an electronic mechanical moment does not
have a value greater than several units, so that the limiting transition
to classical theory cannot be performed. An atom can also have a
magnetic moment (as opposed to an electric dipole moment). There
fore, let us determine the magnetic susceptibility arising from the
orientation of atomic magnetic moments in an external field H.
Let us suppose that an atom in the ground state possesses an orbi
tal angular momentum L, a spin angular momentum S and a total
angular momentum J. In other words, the ground state is a multiplet
state. Let the multiplet splitting (fine structure) be defined by the
Sec. 40] BOLTZMANN STATISTICS 445
energy interval A, so that the level with the closest value J I differs
from the ground level by the quantity A. If the energy of the ground
level is e , then the closest level has an energy e + A. The ratio of the
number of atoms in the ground state to the number of atoms in the
closest state, belonging 1 to the given multiplet, is, according to (40.1)
2J+1 (40.49)
1)41
Thus, if the multiplet splitting A is considerably greater than 0,
the majority of the atoms are in the ground state. If they are placed
in an external magnetic field then each of the multiplet levels is
split into 2 J\ 1 levels, corresponding to its value of J. Suppose that
the field corresponds to the anomalous Zeeman effect in the sense
that the splitting of each multiplet level in the magnetic field is con
siderably less than the finestructure splitting A as defined in Sec. 35.
Then, from (35.11), the energy of an atom in the ground state is
where gL is the Lande factor [see (35.12)] and [JL O is the Bohr magneton.
The number of such atoms is given by the Boltzmann distribution
n(J z )=Ae e . (40.51)
We must again determine the mean value of the magnetic moment
projection on the field:
t^H/,1
*
We have put the minus sign on the left because the electronic charge
is negative. Formula (40.52) involves a statistical sum. The summation
is performed only over those levels which are obtained when the ground
state of the multiplet is split in the magnetic field, since the number of
atoms in an excited state is small.
Summing the geometric progression, it is not difficult here to obtain
a general formula similar to the Langevin formula (40.45). But we
shall confine ourselves to the case of a weak field, when the exponen
tial function is expanded in a series. The expansion must be taken
446 STATISTICAL. PHYSICS [Part IV
up to the second term inclusive because the sum of the terms which
are linear in J z is equal to zero :
j
We calculated the sum of J* 2 in Sec. 30 [see (30.27)]. Using the value
for the sum then obtained, we write the required mean moment
thus :
i __ l vfoLH*J(J+ 1) (2J+1)\ _
ity Q 2 j
!, (40.53)
where we have once again neglected terms of higher order in H.
Formula (40.53) is completely analogous to (40.46) for the electric
moment of dipole molecules produced by a field. The characteristic
magnetic moment is represented by the quantity [LQ gL VJ ( J + 1)
so that the Lande factor gL takes into account the spin magnetic
anomaly. Thus, magnetic susceptibility can be calculated from data
obtained from spectroscopic observations.
Paramagnetisin of rare earths. There are almost no elements for
which we can completely verify formula (40.53) as applied to the
gaseous state. But in rare earths the moment of the electronic cloud
is due to the 4 /shell, which occurs, as mentioned in Sec. 33, deep
inside the atom. When such an atom is part of a crystal lattice, the
4 /shell is but slightly subjected to the action of the electric field
of the neighbouring atoms so that its state may be regarded as being
almost the same as for a free atom of a rareearth element. Therefore,
(40.53) is applicable to those chemical compounds of rareearth ele
ments where other elements do not possess a characteristic magnetic
moment. Its agreement with experiment is very satisfactory for almost
all the elements of the rare earth group.
Exercises
1) Find the mean relative velocity of two molecules of different gases occur
ring in a mixture.
Tho relative velocity distribution is given by a formula similar to the t>u
distribution but written for all three velocity components. This formula is
similar to (40.12), but it involves the reduced mass m =  instead
ra x f m 2
of the mass of a single molecule. Hence, like (40.14), the mean relative velocity
turns out equal to
Sec. 41] BOLTZMANN STATISTICS 447
TCW
If the molecules are identical, their mean relative velocity is \/2 times the
mean absolute velocity.
2) Calculate the velocity of a bimolecular reaction r' t if the effective cross
section depends upon the velocity component (along the lino joining the nuclei)
in the following way:
Then, from the general formula (40.28), we find
_ A
The decisive quantity in this result is the exponential factor e . The
quantity A is called the activation energy. It is equal to the height of the poten
tial barrier over which the colliding particles must pass in order that the reaction
may occur. Unlike a thermonuclear reaction, it is assumed here that the motion
Df the reacting particles is classical. Transitions below the barrier make a vanish 
ingly small contribution in chemical reactions.
Sec. 41. Boltzmann Statistics
(Yibrational and Rotational Molecular Motion)
Molecular energy levels. In order to apply statistics to gases consist
ing of molecules, we must classify the energy levels of the molecules.
The fact that a nucleus is considerably heavier than an electron, and,
therefore, moves much slower, is very helpful here. We have used this
in Sec. 33, when considering the question of the binding energy of
two hydrogen atoms in a hydrogen molecule. The eigenf unction
can be found for any relative positions of the nuclei. In a diatomic
molecule the position of the nuclei is defined by a single parameter
the distance between them. The energy eigenvalue of the electrons
depends upon this distance. Adding the energy of Coulomb repulsion of
the nuclei to the electron energy, we obtain, for a given electron wave
function, the energy of the molecule as a function of the distance be
tween the nuclei. For example, in a hydrogen molecule, the curves for
this relationship are of different form in the case of parallel and anti
parallel spin orientations (Fig. 47). The lower curve refers to the state
with a symmetric spatial wave function and antiparallel spins, while
the upper curve relates to the states with an antisymmetric spatial
448 STATISTICAL PHYSICS [Part IV
function and parallel spins. The lower curve has a minimum at r = r e ,
so that hydrogen atoms may form a molecule only in a definite elec
tron state.
In the general case, several different electronic states can have a
minimum. The distances between corresponding potentional curves
are defined from a wave equation of the type (33.23). In this equation
we can neglect terms involving the masses of the nuclei in the denomi
nator. Therefore, the energy scale separating different electronic states
of the molecules is the same as for an atom, i.e., from one to ten elec
tronvolts.
Close to the minimum of potential energy, nuclei may perform small
oscillations. To a first approximation, these oscillations are harmonic
so that their energy is given by the general formula (26.21):
(41.1)
Here, v is called the vibrational quantum number of the molecule.
This number is, naturally, integral. Fig. 47 shows a more general
dependence of energy upon v, taking into account that the potential
energy curve is not a parabola as in Fig. 41. However, practically, the
deviations from formula (41.1) affect but little the statistical
quantities, because dissociation occurs when oscillations with
large v are excited (see Sec. 51).
The frequency to depends upon the electronic state in
nuclear oscillations occur. In accordance with the
general formulae for frequency (7. 10) (7. 12),
co = / \7r*] 9 so that the fre
\ m \ d r z lr = r e '
queiicy is inversely proportional to the square
root of the reduced mass of the nuclei. There
fore, the vibrational quantum is considerably
less than the distance between electronic lev
Fig. 47 els, which is independent of the nuclear mass.
It is of the order of tenths of an electronvolt.
In addition to vibrational motion, a molecule with two atoms may
also perform rotational motion, flotation is most simply taken into
account when the resultant spin of the electrons is equal to zero and
the total orbital angularmomentum projection of the electrons on
a line joining the nuclei is also equal to zero. These conditions are satis
fied in the electronic ground state for nearly all molecules, with the
exception of 2 , the resultant spin of which is equal to unity (but the
projection of the electronic moment on the axis is zero), and NO,
where the spin is one half (and the orbital angularmomentum pro
jection of the electrons on the axis is zero). Disregarding these excep
tions, we may consider a molecule of two atoms as a solid rotator,
i.e., a system of two point masses at a fixed distance r e corresponding
SOC. 41] BOLTZMANN STATISTICS 449
to the minimum of the lower potential curve in Fig. 47 (see exercise 2,
Sec. 30); (our case corresponds to J 3 and k = Q, so that the closest
excited level hi k with k == 1 is moved to infinity. The rotational
moment of the rotator is perpendicular to the line joining the nuclei
since its projection on this line is equal to zero).
As we know from Sec. 5 [see (5.6)] the rotational energy of two
particles is
where m is the reduced mass and mr* is equal to the moment of inertia
of the rotator / 1 . Going over to the quantum formula, we substitute)
the angularmomentum eigenvalue. It is usual to denote it by the
letter K, so that
This formula corresponds to the energy of a symmetric top with
=0 (cf. exercise 2, Sec. 30). It involves the mass of the nuclei in the
denominator. Therefore, the distances between neighbouring rotation
al levels are of the order of a thousandth of an electron volt and less.
Thus, to a good approximation, the total energy of a twoatom
molecule can be written in the form of a sum with three terms :
, / / , l\ t h z K(K\\) , t . ox
s = e t + e v \ z r  s e + &<o iv + A +  ., ^ , (41.3)
where s e ~ (i.e., it is independent of the nuclear mass w), e^^ ,
ni ' ifYi \i '
Zi^*
m
The excitation of electronic levels. If we substitute the expression
(41.3) in the Boltzmann distribution, the latter separates into the prod
uct of three distributions involving electronic, rotational, and vibra
tional states. Let us suppose that a gas is considered with temperature
not exceeding several thousand degrees, for example, 2,0003,000.
Then if the energy of electronic excitation is several electronvolts
(1 ev = l 1,600, since the temperature can be defined in energy units),
the fraction of molecules in excited electronic states is a very small
t'
number: e e . In those cases when there are very low electron levels,
the Boltzmann factor may also be other than a small quantity. But,
as a rule, dissociation of the molecules sets in earlier than excitation
of then electronic levels (see Sec. 51).
Excitation ol vibrational levels. Let us examine the vibrational states.
For generality we may consider not only molecules with two atoms,
but also polyatomic molecules. If the oscillations of such molecules
are harmonic, we can make the transition to normal coordinates, as
was shown in Sec. 7. Then the vibrational energy assumes the form of
29  0060
450 STATISTICAL PHYSICS [Part IV
a sum of the energies of independent harmonic oscillators. The energy
levels for each such harmonic oscillator are given by a formula of the
form (41.1) with a frequency co corresponding to a given normal oscilla
tion.
Molecular oscillations are basically divided into two types: "valent,"
in which the distances between neighbouring nuclei mutually change,
and "deformational," where only the angles between the "valence
directions" change. For example, in a C0 2 molecule, having a straight
line equilibrium form = = 0, valent oscillations alter the distance
between the carbon nucleus and the oxygen nuclei, while deformation
al oscillations move the C nucleus out of the straightline configura
tion. The frequencies of deformational vibrations are several times
less than those of valent oscillations. The estimation 7&co~0.1ev
related to valent oscillations.
In any case, if the vibrational energy breaks up into the sum of
energies of separate independent oscillations, then the distribution
function also splits into the product of distribution functions for each
separate oscillation.
Let us calculate the mean energy for one normal oscillation:
(41.4)
here we have used the same transformation as in the derivation of
(40.43) and (40.52). Formula (41.4) involves the statistical sum for a
harmonic oscillator. The sum of the geometric progression inside the
logarithm sign is very easily calculated. Indeed,
= e 2,{ e
iO
le
Substituting this in (41.4) and differentiating, we get
l
The first term in (41.6) simply denotes the zero energy of an oscilla
tion of given frequency. The oscillation possesses this energy at abso
lute zero because then the second term in (41.6) does not contribute
Sec. 41] BOLTZMANN STATISTICS 451
anything. The second term has a very simple meaning. If we write
the mean energy in terms of the mean vibrational quantum number ~v
(41.7)
then it is obvious that
"=J (418)
(ftw \l
e i) signifies the mean number of
quanta possessed by a vibration at a temperature Q = kT. At a low
temperature,!; is close to zero. For example, for oxygen and nitrogen,
/ico is about 0.2 ev, or 2,0003,000. Therefore, at room temperature
oxygen and nitrogen occur in the ground vibratioual state. In the
case of hydrogen the reduced mass of a molecule is 14 times less than
that of nitrogen. Its vibrational quantum is close to 6,000. In poly
atomic molecules, where deformational vibrations occur, such oscilla
tions can be excited at temperatures of the order of .300600.
Yibrational energy at high temperatures. If the temperature is very
/i6>
high compared with h to, then e e can be replaced by the expansion
1 + ". Substituting this in (41.6), we obtain
The first term does not relate to thermal excitation. Besides, it is
considerably less than 0. Thus it turns out that at a sufficiently high
temperature the mean energy per oscillation is equal to 6, irrespective
of the frequency. The same can be obtained by proceeding from the
noiiquantized expression for the energy of a harmonic oscillator:
p 2 , ra<o 2 q 2 , A ~ irk .
e ^ +  < 4L10 >
Substituting this in the Boltzmann distribution and calculating the
mean energy, we have
<x> co e a
(j6
' CO 00
J dp j dq
^_~_ .^ ._ = 6 2 ^ln J dp j
j dp j dqe
CO  OO
oo  oo
The statistical integral inside the logarithm is calculated in the usual
way:
29*
452 STATISTICAL PHYSICS i [Part IV

__ _ _ __ . 
f e 8 dp e * d? = V27ue.y^= 6. (41.12)
J Z J K W<0* to
 oo oo
Whence ew=0. Then the total vibrational energy of a gas occurring
at a frequency G> is
''kT, (41.13)
and its contribution to the specific heat is correspondingly equal to R
[see (40.17)]. Thus, at a high temperature the specific heat due to
vibrational degrees of freedom tends to a constant limit.
The excitation of rotational levels.* Let us now consider rotational
energy. The weight of a state with a given value of moment K is,
as usual, equal to 2 K + 1, in accordance with the number of possible
projections of K ' . Especially interesting is the case when a diatomic
molecule consists of two identical nuclei. In classifying the states of
such a molecule it is necessary to take nuclear spin into account.
Indeed, the wave equation for a molecule consisting of identical atoms
does not change form when the nuclei are interchanged. Therefore,
if the nuclei have halfintegral spin, the wave function must be anti
symmetric with respect to the interchange of both nuclei, while if
they have integral spin it must be symmetric. The symmetry of the
eigenfu notion of a molecule is determined by the symmetry of its
factors fin the approximation (41.3) it is separated into factors]:
electronic, vibrational, rotational, and nuclear spin. The electronic
term of most molecules does not change when the nuclei are inter
changed. The vibrational function depends only upon the absolute
value of the distance between the nuclei and therefore does not change
either. The rotational eigenfunctioii is even with respect to this
permutation in the case of even K, and odd for odd K . Therefore, if
the nuclear spin is halfintegral, then the spin function must be anti
symmetric for even K and symmetric for odd K , so that the resultant
wave function may always be antisymmetric. If the nuclear spin is
integral, the position is reversed, and if it is equal to zero, then odd K
are in general excluded because then the spin factor simply does not
exist.
Rotational energy of para and orthohydrogen. We shall now consider
the rotational states of a hydrogen molecule. The total nuclear spin
for hydrogen can equal unity (the orthostate) and zero (the para
state). The weight of a state with spin 1 is 3 and that with spin is 1.
The state with K is even in the rotational wave function. Hence,
it must be odd in the spin function, i.e., it must have spin 0, (see
Sec. 33). But the state with zero moment possesses the least rotational
* The hypothesis that the rotation of molecules participates in the thermal
motion of a gas was put forward by M. V. Lomoriosov as far back as 1745.
Sec. 41] . BOLTZMANN STATISTICS 453
energy. Therefore, only parahydrogen is stable close to absolute
zero.
At a temperature other than zero all those states, for which the
fr a K (K I 1)
"
Boltzmann factor e 2 " mf l e is of the order unity, are excited. Taking
the moment of inertia of a hydrogen molecule to be equal to 0.45 x
X lO" 40 , we can see that already at ? 7 ^300 K the summation over
all odd moments
h*K(K ( 1)
' " *""
differs from the summation over even moments by several thousandths.
But since the states with even moments are, for hydrogen, nuclear
spiii orthostates, each state with even moment has an additional
weight factor 3 according to the number of projections of spin 1.
Thus, at room temperature, 3/4 of hydrogen is orthohydrogen and
1/4 parahydrogen. If hydrogen is rapidly cooled the ratio 3:1 is
retained for a long time because the orthoparatransition proceeds
slowly. Such a state is obviously iionequilibrium since all the hydro
gen in an equilibrium state, at a temperature close to absolute zero,
must be in the parastate.
One of the methods of obtaining pure parahydrogen is to adsorb
hydrogen onto any substance that disrupts the molecular bonds
during adsorption, for example, activated carbon. When desorbing
the hydrogen by pressure reduction at low temperature, the change is
that to the parastate. If the hydrogen is then heated to room temper
ature it stays in the parastate for quite a long time.
Let us now write down the formulae for the mean rotational energy
of ortho and parahydrogen. For simplicity we shall denote the
70 *
factor 3 in the rotational energy by the letter B. Then
B
l)e BK(K + ]
K>0,2,4... _____
__j?i
Y 1 (?. W4n
KO.2,4, ..
K0,2,4,
The difference between "i" or tho an d para is that the summation is
performed over odd K. For a mixture at room temperature
1 3
r~~ "4 para+ 4 "Sorlho (41.15)
454 STATISTICAL PHYSICS [Part IV
At very low temperature, it is sufficient to retain only the term with
K 2 in the summation (41.14), so that
L + 5 e o ^30jSe . (41.16)
For orthohydrogen we obtain
2B
" ~~ + 7e~
2tf 12 7J
6 "T" , S4 e / _i??\
^ # _^ ^2J3\1 f 14 e /. (41.17)
3e~ e
The determination of nuclear spins from rotational specific heat.
The rotational specific heat of hydrogen makes it possible to determine
the spin of a proton. Let us consider formula (41.17). In it, the first
term is a constant. It is due to the fact that a molecule of ortho
hydrogen would have a rotational energy 2 B even at absolute zero.
This energy does not contribute to the specific heat because it does
not depend upon the temperature. Defining specific heat as the deriv
ative T^r , we see that for a sufficiently low temperature the ratio of
the specific heats of ortho to parahydrogen tends to zero as
\B
Therefore, if ordinary hydrogen is rapidly cooled to a low temperature,
its rotational specific heat will be determined by a quarter of its mole
cules in the parastate. It will be four times less than the rotational
specific heat of pure parahydrogen at the same temperature.
Thus, by measuring the specific heat of the equilibrium state of
hydrogen at low temperature (i.e., the parastate) and of rapidly
cooled hydrogen, we can determine the spin of a proton or, knowing
the spin from other data, we can show that protons are subject to
Pauli exclusion because they possess an antisymmetric wave function.
The rotational specific heat of molecules consisting of different atoms.
Diatomic molecules that do not consist of identical atoms possess
equal nuclearspin weights for states with odd and even K. Therefore,
their mean rotational energy is expressed thus:
1)1
(41.18)
The sum inside the logarithm cannot be written in finite form, but
it is easily tabulated. Let us evaluate the temperature at which use
8ec. 41] BOLTZMANN STATISTICS 455
of an integral as a substitute for the summation is justified. Thus,
for hydrogen
B _ J^_ 1.11 10" ! 2 . 10 i4 erg
^ " 2mr? = 1. 67 10 24 (0.74) 2  10" "~ 1 '  1U erg)
which corresponds to a temperature of 87 K.
Here, m is the reduced mass of two protons, equal to half the proton
mass; r e ^0.74 x 10~ 8 cm (where we obtained the moment of inertia
used above). For other gases B is of the order of several degrees so
that for all temperatures at which these gases are not in the liquid
state the ratio J5/0 is a small quantity. To a good approximation, the
summation in (41.18) may be replaced by an integral. If we take
K(K+ 1) = *,
then
(2K+ })dK= 2K+ 1 =dx (dK^ 1)
and
BK(Kf 1) _B*
JT(2/v + l)e "~ e ^je Q dx=~. (41.19)
o
Substituting this in (41.18), we have an expression for the rotational
energy of a diatomic molecule or any linear molecule
/?T
e,e = ^. (41.20)
We note that the concepts of "high" temperature for vibrations and
rotations do not coincide in the least. With respect to the rotational
specific heat of oxygen, the temperature must be higher than 10 K
to be regarded as high, while with respect to vibrational specific heat,
it must be above 2,000 K. Therefore, in a very wide range of temper
atures, in particular at room temperature, the specific heats of diatom 
3
ic gases are constant, and consist of a translational part ^ R and
re
a rotational part equal to B, so that the total specific heat is 5 E.
It may be seen by numerical computation that the rotational specif
ic heat does not tend to a constant limit monotonically, but passes
through a maximum at 6 = 0.81 J3, equal to 1.1 R.
The rotational energy for a polyatomic molecule will be calculated
in Sec. 47.
Exercise
Find the rotational energy of para and ortho deuterium.
Particles with integral spin have a symmetric wave function. Let us now
consider a system of two particles with integral spin, for example, a deuterium
molecule. For comparison we shall also take two particles with spin zero. The
spin function of the latter is identically equal to unity; therefore their orbital
456 STATISTICAL PHYSICS [Part IV
wave function can bo only symmetrical. With respect to the rotational function,
interchange of the nuclei is equivalent to a reflection at the coordinate origin.
Hence, if the spin of a dautoroii were zero then the spectrum of molecular deute
rium would show the linos, corresponding to odd rotational quantum numbers,
to bo absent. In actual fact they exist in the deuterium spectrum, and the weight
of states with even K is twice as great as for those with odd K. This is seen from
the relative intensity of spectral lines that correspond to transitions from the
appropriate states.
We shall show that for a douteron spin of unity, the weight of the ortho
states turns out twice the weight of the parastates. A spin projection of unity
takes on three values: 1, 0, 1. We denote the spin wave functions (of both
deuterons) that correspond to these projections as ^ (1), ^ (0), ^i ( 1) and
^2 (1) ^2 (0) 4*2 (1) Let us form all the spin wave functions of deuterium that
correspond to a total spin projection 0; wo shall only take symmetric and anti
symmetric combinations :
Symmetric functions Antisymmetric f 'unctions
<!/, (1) ^ (" 1) ! *i ( 1) *a (!) > *i (1) *2 ( 1) ~ *i ( 1) ^2 (1) 
For the total spin projection J : 1, we obtain
*i (I) *a (0) I *, (0) * a (1) , 4>i (1) * 2 (0)  *! (0) + a (1) ,
*i ( 1) *a (0) + *i (0) 4 2 (" 1) , *i ( 1) *i (<>)  *, ( 1) +! (0) .
And for a total projection 2 wo have
The symmetric state has a maximum spin projection of two. Hence, the
state for which the spins are parallel is symmetric. But there are six symmetric
spin wave function projections in all, and. spin 2 has 22 + 1 = 5 projections.
Hence, of the functions with zero resultant projection, we can construct one
function corresponding to a zero projection of spin 2. The other function with
zero resultant projection corresponds to a resultant spin 0.
In all, deuterium has six orthostates with a syrnmotric spin wave function.
A spin unity has states given by an antisymmetric spin function because the
maximum spin projection in these states is equal to unity. Thus, there are three
parastates. An even rotational function of a deuterium molecule corresponds
to the orthostates, and an odd rotational function corresponds to the para
states. Then the total function is symmetric, as the case should be for integral
particle spins. The weight (duo to spin) for the orthostates is six and for the
parastates it is three. Therefore, the statistical sum of ortho deuterium is
BK(K+l)
K 0,2,4,...
and for paradouterium it is equal to
3^ (2^+1)
K =1,3,5,...
Sec. 42] THE APPLICATION OF STATISTICS 457
Here the equilibrium state at absolute zero is the orthostate. The energies
of both states [see (41.16) and (41.17)] are
_ 6 ?
"o" .
Compared witli hydrogen, the ortho and parastates are interchanged here.
(lose to absolute zero, the basic contribution to the specific heat is given only
by the orthostate. Two thirds of all the molecules in equilibrium deuterium
occur in this state at room temperature. Therefore, the rotational specific heat
of rapidly cooled deuterium is less than that of equilibrium deuterium at the
same temperature in the ratio 2/3. Thus, by measuring this ratio we can show
that the spin of a deuteron is equal to unity and not zero.
Sec. 42. The Application of Statistics to the Electromagnetic Field
and to Crystalline Bodies
The statistical equilibrium o matter and radiation. In this section
we shall first of all consider radiation in a state of statistical equilibrium
with matter. The conditions for such equilibrium are achieved inside
a closed cavity in an opaque body. The walls of the opaque cavity
absorb radiation of all frequencies and hence they also radiate all
frequencies: if a direct quantum transition is permissible, then the
reverse transition is also permissible. Therefore, radiation arrives
at a statistical equilibrium with matter, that is, in unit time there
is an equal amount of absorbed and emitted energy of electromagnetic
radiation per unit surface of the cavity for every direction, frequency,
and polarization.
An equilibrium density of radiation energy is thus set up in the
cavity. It can be shown that in this case temperature of radiation
is equal to the temperature of the walls. The necessity of this will
be especially clearly seen in the sections dealing with the fundamentals
of thermodynamics (Sec. 45 and 46); for the time being we shall
merely note that it is natural to regard the temperatures of systems
in equilibrium as identical.
The absolutely black body. Equilibrium radiation can be experi
mentally studied by making a small aperture in the wall of the cavity :
if it is of sufficiently small dimensions the equilibrium state will not
be noticeably changed. Radiation incident on such an aperture
from outside the cavity is absorbed in it and does not get outside.
In this sense the aperture resembles a black body which does not
reflect light rays. For this reason it is called an "absolute black body''
and the equilibrium radiation coming from the aperture is called
"blackbody radiation."
468 STATISTICAL PHYSICS [Part IV
This term is somewhat paradoxical since it contradicts the obvious
picture. Indeed, an absolutely black body in equilibrium radiates
more than a nonblack body because it absorbs more, and in equilib
rium the radiation and absorption are equal. If a body having a
cavity and aperture is brought to an incandescent state, the aperture
will exhibit the brightest glow.
The statistics of an oscillator field representation. Planck's formula.
In this section we shall consider the application of statistics to
equilibrium radiation. For this it is necessary to quantize the radiation.
Unlike the statistics of a gas, the statistics of radiation does not
permit a limiting transition to equations, with the quantum of ac
tion being eliminated entirely. This will become clear a little later.
In quantizing the field, a double approach is possible. Firstly,
a field may be represented as a set of linear harmonic oscillators
by characterizing each oscillator with a definite wave vector k and
polarization a (or~l, 2). It is obvious that all these oscillators are
different (as to their k and cy). The quantum properties of such os
cillators are not apparent in calculating the number of states of the
field; their only manifestation is that the energy of each of them
cannot be equated to an arbitrary number, but belongs to an oscil
latorenergy spectrum; i.e., equal toAco \n + ^1 where n is an integer.
When an oscillator is in thermal equilibrium, the mean number
of its vibrational quanta is given by a formula similar to (41.8):
h to
~ 6 ~
(42.1)
v '
The energy of each quantum is equal to Aco and the number of oscil
lations with frequency o> is, according to (25.24),
(422)
Here, in contrast to formula (25.24), both possible polarizations
of oscillation with a given frequency are taken into account, and
K = co/c has been substituted. Hence the energy of an electromagnetic
field in the frequency interval dco is
\ C0 o> t AC\ e*\
to) =  jg  r . (42.3)
The radiation spectrum of the sun is close to this frequency distri
bution.
The statistics of light quanta. Let us now approach formula (42.1)
from another direction. We have said that the electromagnetic
field is viewed as an assemblage of elementary particles light quanta.
Sec. 42] THE APPLICATION OP STATISTICS 459
Quanta of the same frequency, direction, and polarization are in
distinguishable from one another. Therefore quantum statistics
are applicable to them as to particles. At the same time quanta
have integral angular momenta; this was mentioned in Sec. 34.
Therefore they are not subject to Pauli exclusion, and possess a Bose
and not Fermi distribution. But, as opposed to gas molecules, which
are subject to a Bose distribution, the number of quanta is not a
constant quantity, since quanta may be absorbed and radiated.
This is why the supplementary condition (39.12) does not apply
to quanta.
It is easy to pass from the general Bose distribution to a special
case, when condition (39.12) is not imposed; for this it is sufficient
to put equal to zero the parameter (JL, by which equation (39.12)
is multiplied ((ji was introduced to satisfy the 1 condition N = const).
Then the Bose distribution is simplified:
(42.4)
Taking into account that for a quantum s = Aa>, we once again
obtain (42.1). Thus, formula (42.1) denotes either the mean vibra
tional quantum number of an oscillator in an assembly subject
to Boltzmann statistics, or the mean number of light quanta subject
to Bose statistics. As we have already said, certain oscillators obey
Boltzmann statistics: they are differentiated by the numbers n l9
n 2> n & CT ( see Sec. 27), while the statistics of distinguishable particles
is nonquantum. Let it be recalled that we differentiate between
quantum and nonquantum statistics according as the particles are
distinguishable or not.
The impossibility of the limiting transition h > in the statistics
o! the electromagnetic field. Let us now turn, for a time, to the oscil
lator picture. On classical theory, the mean energy of an oscillator
is equal to [see (41.11)(41.13)]. If we multiply it by dg (co), the
classical Ray leigh Jeans formula for the energy of equilibrium
radiation results.
^(coUss==^~0. (42.5)
But this formula is obviously inadequate for large frequencies:
upon integration with respect to co it gives an infinite total energy.
It was precisely here, in statistics, that the classical representations
first so obviously failed. Therefore, in 1900, Planck proposed for
mula (42.3): it was here that the quantum of action appeared for the
first time in physics.
Formula (42.5) is correct only for frequencies that satisfy the
inequality
460 STATISTICAL PHYSICS [Part IV
The total energy of equilibrium radiation. It is easy to find the
total energy of equilibrium electromagnetic radiation from formula
(42.3). Integrating with respect to co, we obtain
CO CO
Vh /" co 3 rf<o Vh
h I" w 3
^ I /,<>
J e''
*L ( **. (42 6)
/** J e*l ' ( >
The integral in (42.6) is merely an abstract number, equal topr
(see Appendix, p. 586), so that the required energy is proportional
to the fourth power of the absolute temperature (the Stefan Boltz
mann law).
Radiation from an absolutely black body. The result (42.6) can
be verified from the emissivity of an "absolutely black body." It
is easy to relate it to the energy S\ For this it is sufficient to calculate
how many quanta fall from inside in unit time upon unit surface
of a cavity, normal to the surface. We have indicated that if we
take away a small section of the wall, radiation will pass through
the aperture with the same composition as that falling on the wall.
The velocity of each quantum is c, so that its normal component
is equal to c cos ft, where ft is the angle with the normal. In unit
time these quanta will strike a square centimetre of the wall from
the whole volume of a cylinder with base 1 cm 2 and height c cos ft.
The energy included in the volume of this cylinder is equal to ^r c cos ft .
The fraction of quanta flying in unit solid angle is equal to ~ ,
so that the total energy falling on a square centimetre of the wall
in unit time is
27T 2
P 4 . (42.7)
The constant in front of T 4 is equal to 5.67 x 10~ 5 erg/cm 2 sec deg 4 .
Formula (42.7) cannot be directly applied to an incandescent
solid body without ascertaining to what extent it may be regarded
as black.
Due to the fact that the sun's luminous shell (chromosphere) is
nearly opaque to radiation, the spectrum it emits is close to the
equilibrium spectrum (42.3), even though it does not exactly coin
cide with it. The temperature of the chromosphere, as determined
from (42.3), is approximately 5,700.
The pressure of equilibrium radiation. It is also easy to calculate
the pressure of equilibrium radiation. It is convenient in doing so
to apply the same reasoning that led to formula (42.7). Now, however,
Sec. 42] THE APPLICATION OF STATISTICS 461
instead of calculating the number of quanta, it is necessary to cal
culate their normal component of momentum transmitted through
a square centimetre of surface. This component is equal to the quan
tum energy Aco divided by c and multiplied by cos ft. Therefore,
unlike formula (42.7), we must integrate cos 2 & instead of cos fr.
In addition, for every incident quantum in the equilibrium state
there is a similar quantum radiated in the reverse direction, so that
the transferred momentum is doubled. Whence the pressure is
2rc 7t/2
= w ( 42  8 )
i.e., one third of the energy density. The same would be obtained
from the derivation of equation (40.22) if the momentum were put
equal to s/c instead of mv. We note that in Lebedev's experiments,
where the pressure of a directed beam was measured, and not of
iight arriving uniformly from all directions, p = <g'/V' 9 the pressure
of the directed beam is equal to the energy density without the
factor 1/3 (see Sec. 17).
From (42.8) and (42.6), the pressure of electromagnetic radiation
increases in proportion to the fourth power of the temperature while
the gas pressure is proportional to the first power. Therefore, radiation
pressure will always predominate at a sufficiently high temperature.
At high temperatures the pressure of a substance can always be
calculated from the ideal gas formula, because the interaction energy
between particles becomes small compared with their kinetic energy.
Hence,
JVO
By considering that atoms are dissociated into nuclei and electrons,
it is easy to express the ratio N/ V in terms of the mass density. Let
us suppose that the substance consists of hydrogen. Then for every
proton there is one electron. If the density of the substance is p,
then the ratio N/V is 2p/m, where m is the mass of a proton and
the factor 2 takes into account the electron. This gives
(42.9)
From (42.8) and (42.6), the radiation pressure p r is
From this we obtain the relationship between density and tem
perature when the radiation pressure becomes equal to the gas pres
sure:
462
STATISTICAL PHYSICS
[Part IV
6 3 =1.5xlO 23 T 3 .
r~~ 90 (he)
For example, for a density p = 1 gm/cm 3 , both pressures become
equal if the temperature is equal to 4 x 10 7 deg. Radiation pressure
is important in the interiors of certain classes of stars.
The frequency corresponding to the maximum radiationenergy
density in a spectral interval rfco. The maximum energy in the distri
bution occurs at a frequency determined from the equation
(42.11)
Performing the differentiation, we have
1e 
30
This equation has a single solution with respect to ~
= 2.822.
(42.12)
(42.13)
16
14
W\
18
0.4
Thus, the frequency corresponding to maximum energy in the spec
trum of blackbody radiation is directly proportional to the absolute
temperature (Wien's law):
Wo== jypl. (42.14)
We notice that the numerical coefficient in
the formula would have been different if we
had considered the wavelength distribution
instead of frequency distribution (see exer
cise 1). Tt is interesting to note that the
corresponding wavelength X in the solar
spectrum is very close to that for the maxi
mum sensitivity of the human
eye. The curve of the distribution
_
h co
1234 5678
Fig. 48
e e 1 is shown in Fig. 48.
Spontaneous and forced emis
sion of quanta. At the beginning
of this section we pointed out
that thermal equilibrium between
atoms and radiation is attained in a closed cavity. The presence
of atoms capable of radiation and absorption is necessary in general
in order that the radiation may arrive at equilibrium ; this is because
Sec. 42] THE APPLICATION OP STATISTICS 463
separate oscillators, corresponding to normal oscillations of the
electromagnetic field, are completely independent of one another,
and any initial nonequilibrium distribution is maintained until
there is an exchange of quanta via absorbing atoms.
In Sec. 34 we derived an expression for the probability of light
emission by an atom. According to (34.46), the radiation probability
in unit time is
We shall now consider atoms whicli are in thermal equilibrium with
matter. Let the frequency co 10 satisfy the relationship Aco 10 e t e ,
where s x and e are the energies of two atomic states. In equilibrium,
atoms with energy e x radiate as many quanta with frequency co 10
as are absorbed by atoms with energy e .
In accordance with the principle of detailed balance, the probabili
ties for direct and reverse transitions are connected by the following
relation :
. (42.16)
Indeed, the firstapproximation formula of perturbation theory
(34.29) is applicable to radiation and absorption processes, since the
interaction of matter with radiation may be regarded as weak. From
this formula, the probabilities for the transitions I >0 and ~>1
are, respectively,
! 2 f/o; W Ql = 2jL\J#> ol \* gi . (42.17)
But according to the Hermitian condition (34.15), the squares of
the moduli of the matrix elements  ^ O i I 2 an( ^ I ^10 I 2 are ^ e same s
that if we multiply expressions (42.17) by the weights of the initial
states, the result will be equation (42.16).
The formula for the probability of absorption related to the case
when a single quantum of frequency co 10 existed in the field before
absorption. If there were n (co 10 ) such quanta before absorption, then
it is natural to assume that the probability of absorbing one of them
in unit time is n (<o 10 ) times greater. This assumption is justified in
electromagnetic field quantum theory.
We shall therefore assume the probability of absorbing in unit time
one of the n (co 10 ) identical quanta in the field to be equal to n (co 10 )
{/o^oi ^ n accordance with the principle of detailed balance we must
have the same probability for the reverse transition, i. e., the emission
of a quantum by an atom occurring in state 1 when there are n (o> 10 ) 1
such quanta in the field ; this is because the transition is reversed with
respect to the one just considered. We represent both transitions
thus:
464 STATISTICAL PHYSICS [Part IV
1st state of the system:
atom with energy e
n quanta with frequency
WIG
quantum
absorption
quantum
radiation
2nd state of the system :
atom with energy c x
n 1 quanta with
frequency co 10
Thus, in accordance with the principle of detailed balance, the prob
ability of emission of a quantum must likewise be ng^V^, which can
also be represented as [(n 1 ) + 1] g^ IF 10 . Because of equation (42.16)
the probabilities for both direct and reverse transitions will be equal.
Hence, if n 1 quanta exist in the field, then the probability of emis
sion is proportional to n, i.e., to the number of quanta increased by
unity. If, for example, there were no quanta in the field before emission,
this factor of proportionality is equal to unity. In this case the emis
sion is termed spontaneous. But when there are quanta in the field,
they stimulate, as it were, further emission of quanta with the same
frequency, direction of propagation, and polarization. The emission
produced by them is called forced. The existence of forced emission
can also be proved by means of quantum field theory, just as the pro
portionality factor n in the absorption probability. The idea of forced
emission was introduced by Einstein.
The derivation of Planck's formula from the relationship between
the quantum emission and absorption probabilities. Let us now consider
atoms in thermal equilibrium with an electromagnetic iield. Let the
quantity n (o> 10 ) denote the equilibrium number of quanta. The condi
tion of statistical equilibrium is that atoms occurring in state absorb
as many quanta with frequency co lo in unit time as are emitted by atoms
in state 1. Then the number n (co 10 ) does not change with time, i.e.,
equilibrium is attained.
The number of acts of absorption by all the atoms in unit time (from
state in which there are N atoms) is equal to
iV )F 01 n(co 10 ). (42.18a)
The number of acts of emission by all atoms in state 1 in unit time is
^i^iohKoJ + l], (42.18b)
because, as we have seen, it involves the number of quanta increased
by unity, i.e., n (co 10 ) + l
Naturally, expressions (42. 18 a) and (42.18b) no longer denote the
probabilities for direct and reverse transitions, but the probabilities
for transitions from a state with the same number of quanta n (<o 10 ),
which probabilities lead to a reduction or to an increase by unity of
the same number. The condition for thermal equilibrium is that these
probabilities are equal:
iV 9 W ol n ( 10 ) = N, W lo [n ( 10 ) + 1] . (42. 1 9)
Sec. 42] THE APPLICATION OF STATISTICS 465
Here, we substitute N and N l from the Boltzmann distribution
(40.1):
<Jo W n n (o> 10 ) = ~ </, W 10 [ (co 10 ) + 1 ] . (42.20)
We now take advantage of the fact that 7i<o 10 e x e , and also
of the relationship (42.16). Then there remains an equation for the
equilibrium number of quanta n (fc> 10 )
n(o> 10 )=(o> 10 )H 1. (42.21)
AVhence Planck's formula is immediately obtained
w (io) = 7r, ( 42  22 )
Thus, the idea of forced emission leads to a correct frequency distri
bution of quanta. Note that the part of forced emission is the more
important the greater n is compared with unity. But large n corre
spond to the classical limit ; it follows that forced or induced emission
is by nature classical and spontaneous emission is a quantum effect.
We notice that the theory of a field consisting of Boso particles
always leads to the concept of forced emission, provided the principle
of detailed balance is used.
The probability of the appearance in the field, of the (n + l)th par
ticle is proportional to n+l, while the probability of the particle dis
appearing is proportional to n. Naturally, if the Bose particles are
charged (as, for example, Timesons) then only those transitions are
possible which are compatible with the conservation of total charge
of the system.
As regards Fermi particles, we must bear in mind that a transition
to a filled level is impossible. Therefore, if the probability that a level
is filled is /, then the number of transitions to this level in unit time is
proportional to 1 /.
The oscillation spectrum for the lattice of a solid body. Let us now
apply statistics to the crystal lattice of a solid body. As applied to
the crystal lattice, statistics is in many ways similar to the theory of
equilibrium radiation.
The vibrations of atoms in a lattice may be described in normal
coordinates, after which their energy is reduced to approximately
the same form as (27.22) : it consists of the sum of the energies of sepa
rate oscillators. To each oscillator, there corresponds a travelling wave
(hi the lattice) of the displacements of atoms from their equilibrium
positions. An example of such a wave, travelling along a chain of
atoms, will be given in exercise 4.
300060
4(5(5 STATISTICAL PHYSICS [Part IV
However, there exist the following differences between the set of
oscillators for an electromagnetic field and those for a solid crystalline
body.
1) The number of degrees of freedom for an electromagnetic field
is infinite, so that it always contains all frequencies from zero to oo.
A solid body has a finite number of degrees of freedom equal to 3A T ,
where N is the number of atoms. Therefore, the range of vibrational
frequencies extends from zero to some maximum frequency co ma x.
2) The dependence of frequency upon the wave vector of an electro
magnetic field is defined by the simple law G>=^C/C. In the oscillations
of a solid body, the frequency depends upon the wave vector in a very
complex manner. Only in the limit do the atomic vibrations become
elastic vibrations of a continuous medium for very long waves (i.e., for
small k), so that the atomic structure of the crystal can be ignored.
In a continuous medium the frequency is proportional to the wave
vector co ^=u a Ic, where the index a must denote that the wave velocity
u depends upon its polarization. Here, as opposed to the electromagnet
ic field in an elastic body, there are three wave polarizations for each h
(when taking into account the atomic structure of a crystal, waves of
another type besides elastic sometimes occur: see Fig. 49). The direc
tions of polarization depend upon the elastic properties of crystals
and upon k.
In an isotropic elastic body two of these polarizations are transverse,
for which the velocity is equal to u t , and one is longitudinal with velo
city ui, so that a ranges through three values as in the crystal.
The number of oscillations occurring in a given interval of values k
can be obtained, as usual, by proceeding from the relationship between
the oscillation number and the wave vector. Each oscillation is defined
by three integers (%, n 2 , n 3 ) and cr. The wave vector has components
proportional to n v n 2 , n^:
; k , = 1^ ; t t ~ 1^ (42.23)
* V '
(a v a 2 , a 3 are the crystal dimensions). From this,
7 j 7 7 a*a.,(t*dk T dk u dk, Vdk x dk u dk f ,*~^j*
rftfk = dn 1 dn a d 3 = l  (27r)3  =  (i) ' (42>24)
Comparing this formula with (25.22), we notice that now the denom
inator is (2 7u) 3 , while before it was simply 7r 3 . The difference is due
to the fact that here the numbers n l9 n 2 , n 3 are found from the peri
odicity condition, as for an electromagnetic field [see (27.4)], and the
expansion is performed for travelling waves instead of standing waves.
For this reason, an additional factor 2 3 appears in the denominator
of (42.24) as compared with (25.18). But here the numbers n lt ra 2 , n 3
range through all values from oo to oo, while in Sec. 25 they varied
Sec. 42] THE APPLICATION OF STATISTICS 467
only from zero to oo, filling one octant. Thus, the total number of
states turns out to be the same, irrespective of the method used to
count them by travelling waves or by standing waves. And this is
the way it should be [cf. (25.23)].
The energy of a solid body. It is now easy to write down an expression
for the energy for an interval dk x , dk Y , dk z and for a given polarization
of oscillations cr. Like in (42.3) we have
d (k, a)  *,dk.dlc,dlc. (42>25)
V ' / , hto a v V '
S^\e l)
In order to find the total crystal energy, we must integrate this
expression over all dk x dk y dk z and sum over CT. Unlike the case of an
electromagnetic field, here the integration must not be performed to
infinity, but only between limits such that the total number of oscil
lations equals the number of degrees of freedom 3iV (because there
are N atoms in the lattice and each one lias three vibrational degrees
of freedom) :
In order to find out what is meant here by summation over a, let us
consider the possible types of vibration of a crystal lattice composed
of atoms.
Two types of lattice vibration. If we confine ourselves only to crystal
lattices of elements where there is only a single atom in an elementary
cell, then the index a will indeed range in value from 1 to 3. In reality,
lattices are sometimes of more complex form, and the possible types
of vibration become correspondingly more complicated. This can be
illustrated by a simple example. Let there be two atoms in a cell,
as shown in Fig. 49 by solid and open circles. The length d corresponds
to one constant of the lattice. Let us imagine a vibration of some
definite wavelength X. In Fig. 49 X = 4 d (a single half wave is shown).
This vibration may be effected in two ways: both atoms in the ele
mentary cell are either displaced to one side (Fig. 49a), or in opposite
directions (Fig. 49b). The
second vibration corre
^ sponds to a greater fre
quency for a given wave
length than the first be
oJ 6) cause the restoring force
Fig. 49 for the second vibration
is greater.
If there are i atoms in an elementary cell then 3 i types of vibration
exist in the threedimensional case. Three types correspond to the
case (a) in Fig. 49, when all the i atoms are displaced in the same direc
30*
468 STATISTICAL PHYSICS [Part IV
tion, and, in the limiting case of long wavelengths, the whole lattice
vibrates like a continuous medium.
The total number of crystal vibrations is equal to 3 iN' 3 A 7 , where
N r is the number of elementary cells. Obviously, the number of vibra
tions is equal to the number of degrees of freedom, i.e., three times
the number of atoms in the lattice.
Calculating the energy of a crystal lattice. (42.25) cannot be integrat
ed in general form because the dependence of frequency upon k
and (j is different for different lattices and for different types of vibra
tion. We must therefore coniine ourselves to two cases.
a) The temperature is considerably greater than the limitingfre
quency quantum 7&6) max . It is then all the more so greater than the
other quanta, so that we can neglect all terms, except the first, in the
exponential series
h co a
e  1 =
Substituting this in (42.25), we obtain a simple expression for the lat
tice energy:
' KOJT f I I </A*</MA> _ 3 , yo = 3 RT (4 L >.27)
o
Here we have made use of the fact that the total number of lattice vibra
tions is equal to the number of its degrees of freedom 3iY. Hence,
the specific heat of the lattice is equal to 37? and is the same for all
elements in molar units. This law is well satisfied for very many ele
ments already at room temperature (the Dulong and Petit law).
Exceptions are, for example, diamond and beryllium, for which the
large frequency <o m ax is due to a relatively small atomic weight, since
_ i
frequency is proportional to M  .
Expression (42.27) fits the general law for the temperature dependence
of vibrational energy at high temperatures (41.13).
Very frequently, in a crystal lattice we can distinguish the molecules
of the substance of which it is formed. We cannot, of course, draw a
really strict distinction between atomic and molecular crystals but,
qualitatively, this distinction is fully meaningful. In molecular crystals
we can separately consider the motion of atoms inside molecules
(purely vibrational, in the given case) and the motion of molecules
as a whole relative to their equilibrium positions in the crystal. The
latter correspond not only to definite centreofmass coordinates of
the molecules in the lattice, but also to certain distinct orientations
in space. Usually all the degrees of freedom of the motion of molecules
in a crystal are vibrational. Solid hydrogen forms an exception where
the molecules rotate almost freely (this rotation is similar to the rota
tion of a pendulum if its total energy is sufficient for transition through
Sec. 42] THE APPLICATION OF STATISTICS 469
its upper position). The frequency spectrum for all vibrations, trans'
lational and rotational, consists of very man} 7 ' dispersion curves with
different cr (according to the number of modes of vibration) and with
its frequency dependent upon the wave vector. This spectrum is
complicated by the vibrations of atoms inside molecules, similar to
case (b) in Fig. 49. Since all the possible vibrations are excited by
temperature increases, the dependence of specific heat upon tempera
ture is of very complex form in molecular crystals.
b) The temperature is considerably less than Aco ma x> Then the factor
/ /lw max \
\ I/
is so small that integration can be taken to infinity without any essen
tial error, because only the small frequencies, for which the quantum
is of the order 0, contribute noticeably, i.e., Aco ~6. For large fre
quencies, the Planck factor \e 1; cancels the contributions of
the corresponding vibrations.
However, in the case of small frequencies the lattice vibrations pass
into the vibrations of a continuous medium, for which vibrations the
frequency is related to the wave vector by the simple formula
7')' fc  (42.28)
The propagation velocity of such waves depends upon the direction
of propagation and upon polarization, but does not depend upon the
absolute value of k. The remaining types of vibration, whose frequency
does not become zero for small k, are not excited at low temperature
since the corresponding quanta are comparable with A(o raax .
It is expedient to transform the volume element dk x dk v dk z to spheri
cal coordinates, i.e., to replace it by the expression k z dk dl, where
dQ, is an element of solid angle for the directions k. Here, in accordance
with what has just been said, the integration with respect to k must
be taken to infinity.
Thus, we obtain a formula for the total crystal energy at low tem
perature
The inner integral is taken in the same way as in the formula for the
energy of an electromagnetic field (42.6), so that
470 STATISTICAL PHYSICS [Part IV
Thus the energy of a crystal lattice is proportional to the fourth power
of the absolute temperature, while the specific heat is proportional to
the third power. This refers to temperatures considerably smaller
than /zcomax.
The Debye interpolation formula. P. Debye the author of the theory
of crystal specific heats at low temperatures which is set out here
proposed an interpolation formula for intermediate temperatures
when the results (42.30) and (42.28) do not hold. The Debye formula
reduces to both these formulae in the limiting cases of high and low
temperatures. The intermediate interval is described qualitatively,
but in certain agreement with experiment. In order to obtain the
Debye formula, we suppose that the law
holds for all k, where u is the usual propagation velocity of elastic
waves. We may even take Ui = u z Ut, u^=ui 9 where Ut and u\ are
the velocities of transverse and longitudinal waves in a given substance
in the polycrystallirie state, which velocities are independent of the
direction of propagation of the wave. We define the upper frequency
limit comax from the condition that the total number of vibrations is
equal to 3^. For this we must go over to spherical coordinates in
(46.26):
(42.31)
a
or, changing to ui, iir, we have
lS7T 2 A r V/
(42.32)
^(V + 
Condition (42.31) is selected so that at high temperatures the correct
law < = 3NQ is automatically obtained. At medium temperatures
^ h co m ax ; k= ^ x is substituted as the upper limit in the integral
(42.29) in place of oo, so that the energy expression has the form
2 , i \ Tr^do (42 33)
e w  1
Changing to the integration variable x = ~ and denoting
^60, we can rewrite the lattice energy thus:
See. 42] THE APPLICATION OF STATISTICS 471
r /j M**.!.
2K a \ u] u*. I h* J e x 1
At low temperature OD > 6, so that the upper limit in the integral is
replaced by infinity. Then the integral is equal to jg , and for the
energy we have
2 ,
The exact formula (42.30) assumes the same form if we replace u a
in it by u t and m, which are independent of direction.
We shall now show how to determine 0D from experimental data on
specific heat and, independently, from elastic constants. The following
values of specific heat G are known for tungsten (from the data of
F. F. Lange) : T  26.2K, C  0.21 cal/mol  deg. ; T = 3S.9K,
C = 0.75 cal/mol deg. The cube of the temperature ratio is equal to
3.37, and the ratio of specific heats is 3.58. We may assume that in
the given temperature range the T 3 law for specific heat holds. Substi
tuting OD = A (Omax in formula (42.35), we determine o) max with the aid
of (42.32). This gives
Converting this to heat units, we write
Here R = 1.96 cal/mol deg. Substituting the specific heat at the lowest
temperature, we find TD 340.
We now determine TD by proceeding from the elastic constants for
tungsten. We have to give, without derivation, the formulae which
connect u t and u\ with the shear modulus and the bulk modulus for
tungsten (see L. D. Landau and E. M. Lifshits, The Mechanics of
Continuous Media, Gostekhizdat, 1953, p. 744 or A. Love. A Treatise
on the Mathematical Theory of Elasticity, Oh. XIII, Cambridge, 1927).
Ut'
F?.
Here, K is the bulk modulus, which, for tungsten, is about 3.14 x 10 12
dyne/cm 2 at low temperature. O is the shear modulus equal to
1.35xl0 12 dyne/cm 2 . The density of tungsten is p = 19.3 gm/cm 2 .
Hence, ui = 6 x 10 5 cm/sec, u t = 2.64 x 10 5 cm/sec. For tungsten the ratio
is equal to 0.635 x 10 23 . Whence, if we calculate it from (42.32),
ix A *i 1AM i Am 4.61 X 10 13 X 1.05 X 10~ 27 Oeo0
equal to 4. 61 x 10 13 sec~ 1 andTD= 7^3 npie =^352.
472 STATISTICAL PHYSICS [Part IV
The agreement with what was obtained from specific heat turns
out to be even better than could have been expected, because the
elastic constants do not strictly refer to the temperature at which
the specific heat was determined, and also because tungsten is a
crystalline substance and its elastic properties are characterized by
three moduli of elasticity instead of two (see Landau and Lifshits,
loc. cit., p. 675). For a number of substances we have the following
values of Debye temperature T D : Pb 88, Na 172, Cu 315,
Fe 453, Be 1,000, diamond J,860 (all from absolute zero).
At high temperature, > OD, we must put ?.* 1 p^ x 9 so that
which is what we demanded.
For * OD, formula (42.34) agrees with experiment qualitatively.
We note that we must not expect complete agreement, because the
initial assumptions made in deriving this formula are not quantitative
in character. It is not worth the attempt to make formula (42.34)
more accurate, without taking into account the exact form of the
dependence of co upon k. The attempts at correcting this formula,
which are sometimes made, are simply in the nature of adjustments.
Exercises
1) Write down the formula for the wavelength distribution of blackbody
radiation energy. Proceeding from the fact that co   . , wo have
' ' / 2nhc
X'U X0 "l
The maximum is defined by the equation
'2) Show that if Hose particles interact with a Boltzmann gas the probability
of a particle appearing in a certain state is proportional to n j 1, where n is the
number of particles already in that state, arid the probability of a particle
disappearing is n.
Let the energy of a Boltzmann particle be e and that of a Bose particle, t\.
Lot us consider the process in which there occurs the transition
e f Y) > e' f YJ'
i.e., the interaction of these particles changes their initial state with energies
e, Y) to a state with energies e' and r{ '. In statistical equilibrium we must observe
the balance
JT r ee ' N e >jr, (1 + ') == TfV.AY (1 I 7iYi)HYj'.
where Wit' is the probability of direct transition and Wt't is the reverse
transition probability. Putting
Sec. 42] THK APPLICATION OF STATISTICS 473
It e He'
^ \i / n' M. X \i
I/ , HV = U I/
wo see that tho balance equation is satisfied if We, &' W&' e. For simplicity
wo have put gs ge.'. The presence of spontaneous emission is duo to the Bose
distribution.
3) Find the total number of quanta in blackbody radiation at a given
temperature
oo on
AT / C0 *** * / ^ _
9 T" I I ~~ > T"' * I ~ v ^1 *
TT 4 c** I _<> IT"/?. c j I e x I
./ e " 1 J
Further (see Appendix),
Hence,
CO CO 00 OO
r 27 /' Z 1 
=10 =1 M 1
The sum is approximately 1.2, so that
_2.4 FO 3
7T 2 "/>C 3 '
4) The atoms are situated in tho form of a linear chain. We shall denote
the displacement of the nth atom by a n . Tho force acting between the nth
and (n + l)th atoms is equal to a (a n + l ). Find the equations for tho vibra
tions of tho chain. Ignore the interaction between the more distant neighbours.
The vibration equation for the nth atom is
md n  a (a n + i f !
We look for a n in the form
Substituting this in the initial equation, we find, after cancelling e^ n
ml (t) = a& (t) (e { f+ e~^ 2)  2 a6 (t) (cos / 1)  4 a sin 2   6 (t) ,
&
so that the oscillation frequency for a given value of / is
sin
If the distance between the atoms is d then n r , where x is the equilib
rium position of the nth atom. Putting j =k,wQ have e ifn = e ikx , so that / can
be called the wave vector, considering that tho length is measured in units
of d. For small /, as was asserted, the frequency is proportional to / :
474 STATISTICAL PHYSICS [Part IV
Sec. 43. Bose Distribution
The choice ol sign of JJL. The Bose distribution has very peculiar
properties at low temperatures. We shall suppose that the atoms
do not have spin; such, for example, are helium atoms with atomic
weight 4. Both the electrons in the cloud of the helium atom and the
protons and neutrons in the helium nucleus are in the Isstate. They
all go in pairs and by the Pauli principle the spins are antiparallel.
Therefore, the resultant spin is zero.
From (39.30), the weight of the state of a spinless particle is
The normalization condition (39.23) looks like
Vm'l,
. .
 ( '
"'"I
This condition can be satisfied only for negative (ju Indeed, if we
suppose that (JL is greater than zero, then the denominator of the
s tx
integrand will be negative for s < (JL because then e e < 1 . But
this is impossible because the distribution function is, by its very
meaning, a positive quantity.
Hence, fji<0. At high temperatures the Bose distribution passes
into the Boltzmann distribution in accord with (40.6).
The sign ol ~ . As the temperature diminishes, [i decreases in
absolute value. This can be shown generally with the aid of (43.2).
Differentiating this equation as an implicit function we have
CO OO
/r
_jrf? _ /
r /
e e 1 J
8 ye </e j e  (x
sir
00'
5
The integrands in (43.3) are essentially positive quantities [(e p.) > 0,
because (Ji<0], and therefore ^~ <0. Hence, as decreases, the
absolute value [JL diminishes monotonically since JJL must increase.
Sec. 43] BOSE DISTRIBUTION 475
We shall now show that (JL becomes zero at a temperature other
than zero. To do this we put [x = in (43.2) and find the corresponding
value 0=0 :
The integral simply represents an abstract quantity: it is equal
to & 2.31 (see Appendix). Therefore equation (43.4) is satisfied by
a value of 6 that is different from zero.
Bose condensation. What will happen when the temperature is
reduced further ? (ji cannot go from negative to positive values since,
as we have shown at the beginning of the section, this would lead
to negative probability values. (JL cannot become negative once again,
because  y is always less than zero so that (JL varies only monotonically,
if it is at all capable of varying. Therefore, the only possibility is
for JL to remain equal to zero after it has once attained its zero value.
But then equation (43.2) is no longer satisfied if the temperature
is less than 6 , and N does not change. On the contrary, it can be
seen from (43.4) that if we define the number of particles as
for 0<0 , it decreases with the temperature in proportion to 0*/a.
What happens to the remaining particles which number N N"i
As opposed to light quanta these particles cannot be absorbed.
Therefore, they will pass into a state which is not taken into account
in the normalizing integral (43.2). The only state of this kind possesses
an energy equal to zero: due to factor A/e it does not contribute
anything to the integral (43.4). In normalization we can isolate the
particles occurring in the zero state in a separate term. If a finite
number of particles go to the zeroenergy state, they will naturally
fall out of the integral. N' particles remain continuously distributed,
but with the value \i = Q. Thus, at a temperature 0<0 , the whole
distribution consists of an infinitely narrow "peak" at e = and of
/ _L \i
particles distributed according a \e e I/ law. At absolute zero
all the particles are in a zero state : this state of a Bose gas is obviously
defined uniquely. It will be noted that a BoJtzmann gas would behave
in an entirely different way when the temperature tended to zero.
476 STATISTICAL PHYSICS [Part IV
Liquid helium. Helium with atomic weight 4 obeys Bose statistics
since the spin of its nuclei and of the electronic shells is equal to zero.
It is therefore interesting to see whether anything like this "Bose
condensation" is observed in helium.
It is difficult to give a unique answer because at low temperature
helium is a liquid, and the Bose distribution, which relates to an ideal
gas, does not apply. Nevertheless the qualitative aspect of the result
obtained for a gas may still hold. Namely, it may be supposed that
at a certain temperature part of the gas will pass into a zero energy
state and, accordingly, will not contribute to the specific heat.
Liquid helium does, in fact, experience a peculiar change of state
at a temperature of 2.10K (at atmospheric pressure). Speaking of
a monatomic liquid, which is what liquid helium is, it is difficult
to imagine any change of state related to a rearrangement of the
atoms in space. Therefore, it is interesting to compare the actual
temperature of transition in liquid helium with the temperature at
which Bose condensation would occur in gaseous helium of the same
density.
The density of liquid helium is equal to 0.12 gm/cm 3 . Whence
the ratio y ' 4  X 6 x 10 23 = 0. 1 8 x 10 23 . Consequently, according to
(43.4) the temperature is
0.18  10 23 9.8G 1.41  1.18  10 8I W, ,. . lf .
2.31.17.1.10 ) /J  3.86 101*;
, _ 3.8(^10 ' _
~ 1.38 ."10 ^' '
which is close to the transition temperature. At the transition, the
specific heat of helium experiences a discontinuity. In the case of
a Bose gas, only the derivative of the specific heat with respect to
temperature has a discontinuity.
Superfluidity. P. L. Kapitsa discovered that below the temperature
of phase transition, liquid helium possesses a most remarkable prop
erty : it is capable of passing through the finest slit without exhibiting
any signs of viscosity. This property was called superfluidity.
L. D. Landau developed a theory of superfluidity proceeding from
the supposed quantumlevel spectrum for a liquid. On the basis
of this theory he built the hydrodynamics of a superfluid, which
differs from conventional hydrodynamics in that each point possesses
two velocities instead of one : a normal and a superfluid component.
The occurrence of two velocities means that in a superfluid two types
of sound vibrations may be propagated: ordinary sound, in which
pressure and density oscillate, and "second sound/' which is connected
with the relative motion of the normal and superfluid components.
The second sound was demonstrated in an experiment carried out
by V. P. Peshkov using a method proposed by E. M. Lifshits. The
Sec. 44] FERMI DISTRIBUTION 477
experimentally found velocity of second sound (which is small com
pared with the velocity of conventional sound) is in excellent agree
ment with Landau's theory.
The question of th$ relationship between superfluidity and Bose
condensation cannot be considered fully resolved. It may be suggested
that the superfiuid component corresponds to that part of the helium
which has passed to the zero state. This hypothesis is strongly sup
ported by the fact that the liquid isotope of helium with atomic
weight 3 is not superfiuid : the nuclear spin of helium 3 is equal to 5 ,
so that its atoms are subject to the statistics of Fermi and not Bose.
Accordingly, they cannot all pass into the zero state together: the
Pauli principle does not permit this.
N. N. Bogolyubov showed that a gas which is close to an ideal
gas and consists of Bose particles possesses an energy spectrum which,
according to Landau's theory, a superfiuid liquid should have. How
ever, no one has so far succeeded in proving theoretically that it is
precisely liquid helium below the transition point that should possess
such a spectrum.
Exercise
Calculate 1 the energy and pressure of a Bose gas below the transition point.
For the energy we have
oo
Vml* 5 , C x*l*d.c 1.78 \ 7 m*l*tflt
g = 4>1  3  /a ^
6
(see Appendix). The pressure is determined from the general relationship
(40.22):
2
Thus, the pressure of a Bose gas below the transition point is independent of
volume and depends only upon the temperature. Jf we compress such a Boso gas
its particles will go to the zeroenergy state. Conversely, upon expansion the
particles will come out of the zero energy state until there are none left. If expan
sion continues the pressure will begin to decrease.
Sec. 44. Fermi Distribution
The form of the Fermidistribution curve and its interpretation.
The criterion for the transition from quantum statistics to classical
statistics is that [see (40.7)]
N ^ #(0) / w
If the inequality is reversed, then essentially quantum properties
of the statistical distribution appear. In this section we shall consider
478 STATISTICAL PHYSICS [Part IV
the properties of the Fermi distribution when the inverse inequality
T" $()* <""
or an equivalent inequality
> 1 (44.2)
is satisfied.
From (39.26) and (39.30), the Fermidistribution curve is of the
following form:
J </e 1 M i Q\
~ , . (44.3)
Here, a weight factor 2 is introduced, since we have put j ' ^ . The
first factor in (44.3) represents the total number of states between
and ( rf, while the second factor represents the probability that
these states are occupied. We can interpret the function
/W= YV  (.4)
e i 1
as a probability and as the mean number of particles, because / (e)
is contained between zero and unity. A similar function in the Bose
distribution could only denote the mean number of particles in one
of the quantum states with a given energy, because the Bosedistri
/ * \ l
bution function \e I/ is sometimes even greater than unity
and must not be interpreted as a probability.
Let us see how the curve / (E) behaves when  g> 1. When =
we obtain
JL
T +i
because e e is a small number. The quantity e e is also a small
number as long as e remains smaller than [JL, while / (e) is close to
en
unity, like / (0). Only when e [i is comparable with 6, is e Q of
the order of unity, so that / (s) begins to decrease noticeably with
further increase of e. For e = (Ji, / (fi) decreases toy:
J_ L
"  e o + 1 " 2
Sec. 44] FERMI DISTRIBUTION 479
For still greater values of e, / (e) decreases exponentially because
unity can then be neglected in the denominator, and, for e > ji, / (e)
becomes the Boltzmann distribution
The Bose distribution also has the same limiting form. The curve
/ (z) is roughly shown in Fig. 50. The region e, where / (s) changes
from unity to zero, has a width of the order 6, since g  is comparable
with unity only if e ji^O: for smaller s the exponential is con
siderably smaller than unity, while for larger e the exponential is
considerably greater than unity.
Fermi distribution at absolute zero. We shall call the region of
transition of / from unity to zero the spread region of Fermi distri
bution. As the temperature decreases the spread
region narrows and, at absolute zero, becomes a
sharp discontinuity /, so that the distribution
function takes the form of a right angle. Fig. 50
shows this step by a broken line. The value of (JL
at absolute zero is called [JL O . Hence,
= 0, all states with energy less
than [JL O arc occupied with unity prob
F . ability (i.e., with certcainty), while
lg ' those with energy greater than [JL O are
empty, also with certainty.
This result can likewise be obtained directly from Pauli's principle
without resorting to statistics. From (39.32), a definite interval of
momentumcomponent values kp x , kp y , &p z corresponds to one state
of particle motion. If the particle is contained in a box with sides
%, a 2 , a 3 , then it follows from the uncertainty relation (23.4) that
since these quantities show by how much the momentum components
of two particles must differ in order that the particles may be regarded
as occurring in different states of motion.
This follows not only from the uncertainty relation, but can also
be seen strictly when computing the states leading to formulae
(25.23) and (39.32). Here, each state must be identified not with
the volume of the parallelepiped, but with one of its vertices whose
coordinates are given by the three integers n l9 w 2 , n 3 . The coefficient 2 it
in the uncertainty relations is taken so that both definitions for the
number of states agree.
If we plot p x , p y , p z on coordinate axes, then to each state of spatial
motion of the electron there correspond three quantum numbers
480
STATISTICAL PHYSICS
[Part IV
n ly n%, n%. These quantum numbers specify the number of the parallele
piped with sides Ap Xy &p y , A^. It is shown in Fig. 51. All the space
in which the axes p Xy p y , p z are drawn can be
filled with such boxes. Since three quantum
numbers correspond to a single box and, in
addition, the state is also given by the spin, there
may be two particles with spin ^ having momen
tum projections in the same interval Ap*, A^ y ,
~ A>*. The spins of these two particles are anti
* parallel.
Thus, the space p xy p y , p z may be divided
into boxes or cells with dimensions
Fig. 51
(44.5)
where there are no more than two particles in each cell.
The closer the cell to the coordinate origin, the less the energy it
1
possesses, because the energy is equal to e
(pl + K
In other words, it is proportional to the square of the distance of
the cell from the origin.
Let us now consider the state of a gas at the absolute zero of temper
ature. If the gas consisted of only two particles, then at absolute zero
the states of both particles would fill the cell closest to the origin.
In accordance with the Pauli jmnciple, the next two particles cannot
enter the same cell : they are forced to take up positions further
from the origin. As the number of particles increases, cells are filled
which are situated further and further from the origin; but each
time two particles are added they fall into a free cell closest to the
origin, because, by definition, absolute zero corresponds to the least
possible energy of the gas as a whole.
If there are very many particles, their cells will densely fill a sphere
whose centre is the coordinate origin. All states inside the sphere
are filled with unity probability, while those outside the sphere are
free also with certainty.
The limiting energy of Fermi distribution. If we denote the energy
corresponding to the boundary of the sphere by e , then it can be
seen from Fig. 50 that Q = \JL O . [i Q is the limiting energy of a particle
at absolute zero. It is very easy to calculate e or [JL O . Since at absolute
zero the function / (e) is equal to unity for all s < [L QJ the total number
of particles N is, from (44.3),
N = ^ " V
F(2m) 3 /2e 3 / 2
(44.6)
Sec. 44] FERMI DISTRIBUTION 481
whence
The same can be seen without the aid of / (s). Indeed, the radius
of the sphere of greatest energy is
Its volume is
But this same quantity is equal to the number of filled elementary
cells (with two particles per cell) multiplied by the volume of a single
cell ~rc Consequently,
4 /rt vy N (2nh)* /AA a\
y7r:(2m )^ ;T"j/ > ( 44  8 )
whence equation (44.7) is again obtained.
At absolute zero the state of a Fermi gas as a whole is defined
uniquely: in quantum statistics it is necessary to indicate which
states are occupied by separate particles, but it is impossible to deter
mine by which particles they are filled. In the given case all the
states inside the sphere with limiting energy s are filled by particles.
The criterion for the closeness of the Fermi distribution to the
distribution at absolute zero (based on the form of the distribution).
At a temperature close to absolute zero thermal excitation can be
imparted only to those particles whose energy is close to e pL .
Indeed, as long as 6<^s , a thermal excitation of the order 6 cannot
be imparted to a particle whose cell lies deep beneath the surface
e = e , because the states between the surface and the given cell
are occupied, and the energy is insufficient to remove the particle
beyond the limits of the surface boundary. Therefore, only those
particles whose energy differs from e by an amount of the order
of can take up free places. Deeper states will remain densely filled
as before. Thus, the filling probability will be almost equal to unity
for all energies <e > an ^ will fall to zero in a region of the order
of close to ^0, as shown in Fig. 50.
The criterion that the curve is close to the step is the inequality
6 < > (44.9)
and this agrees with (44.1) within the accuracy of the numerical
factor. As we shall soon see, the concept of ' 'closeness" of temperature
to .absolute zero, according to the criterion (44.9), differs greatly
from conventional.
31 0060
482 STATISTICAL PHYSICS [Part IV
Electrons conducting electricity in metals are usually considered
as an ideal gas. The main basis for this is the fact that we as yet
have no better theoretical model. It has not been possible to consider
electrostatic interactions between electrons sufficiently fully to obtain
quantitative results that might compare with experiment. This is
why the phrase "electron gas" in metals is used. In many cases the
conclusions from such a model are in good agreement with experiment.
Without considering the electron theory of metals, we shall take
the electron gas only as an example in which condition (44.9) is
satisfied. Let us suppose that there is one conduction electron for
each atom. This assumption appears to be satisfied for alkali metals,
in which the outer electron is weakly bound and is separated from
the atom in a lattice.
Let us find e for the electron gas in metallic sodium. The density
of sodium is 0.97 and atomic weight 23. Hence, unit volume contains
;,?7 6.02  10 23 = 0.25 10 23
Z*j
atoms and as many conduction electrons. Whence, from (44.7),
1 19 10 5 t
e == 2.1 . 4.6 Y 8 .' 10 _ 27 0.08  10 16 = 4.8  10
[the sequence of the numbers is the same as in (44.7)]. In degrees e
is 34,800. Hence, at all temperatures for which we can speak of
sodium as a metal, the electron gas in it is close to a Fermi gas at
absolute zero. Similar results are also obtained for nonalkali metals,
though with a less reliable value of electron density.
The compressibility of alkali metals. Let us derive a formula for
the compressibility of a Fermi gas at absolute zero. From (44.6),
the energy at absolute zero is
e '/.. (44.10)
In accordance with the Bernoulli equation (40.22), the pressure
is equal to two thirds the energy density, i.e.,
7) = ? ( 2m ) /2 e / = Ali!L(i A 3 l*L\ /3 (44 1 1)
^ 15 :T 2 /i 3 5 m \F/ v ;
Whence
 ay =>, ^SrS^)"'' = 273 X 10 27 (^)~ /3 baii. (44.12)
Ya. I. Frenkel noted that the compressibility of alkali metals
is close to the compressibility of an electron gas.
Sec. 44]
FERMI DISTRIBUTION
483
Indeed, expressing N/V in terms of atomic weight and density,
we obtain the following table:
Li
Na
K
Kb
Cs
1 dV
    x 10 8 from equation (44.12)
4.7
13
37
52
79
\ op
I dV
 ^ x 10 6 from experimental data
S
15
32
40
61
In a crystal lattice there are, of course, not only forces of repulsion
between particles, but also cohesive forces. The equilibrium of these
forces with the forces of repulsion determines the characteristic vol
ume which every condensed body, solid, or liquid has in the absence of
external pressure. Ordinary atmospheric pressure gives a force which
is negligibly small compared with these tremendous forces that keep
bodies in their volumes. In order to change the volume of a body by
only one per cent, pressures are required in the order of tons of thou
sands of atmospheres.
The coincidence of theoretical and experimental data indicates that
when alkali metals are compressed the cohesive forces change insignif
icantly compared with the forces of repulsion. It is even conceivable
that the state of the valence electrons in alkali metals is perturbed to a
comparatively small degree by the atomic residues, and, to some ex
tent, is close to an electron gas. Compression affects but little the
electronic shells of the atomic residues, and therefore the compressi
bility of alkali metals is close to the compressibility of an ideal Fermi
gas. That this should bo so is, of course, not at all obvious beforehand.
Paramagrietisni of alkali metals. According to Pauli, the paramagne
tism of alkali metals can also be exp lamed on the basis of the concept
of a free electron gas.
If we place a Fermi gas (consisting of electrons) in a magnetic field,
the energy of the electrons, whose spins are parallel to the field, will
be equal to $H, while the energy of electrons with opposite direc
tion of spin will be equal to p//*. Therefore, if those electrons whose
spin is antiparallel to the field reverse their spin directions, then the
energy of the gas must decrease. But all the places inside the limiting
energy sphere are occupied ; so for an electron to change its spin direc
tion it must come out of the sphere into a free cell. But this increases
its kinetic energy. Equilibrium is established between electrons with
spins parallel and antiparallel to the field when their total energies
become equal. Indeed, if there occurred a further transition of elec
trons into a state with spin parallel to the field, the increase in their
* Here the Bohr magneton is denoted by
confusion with the distribution parameter ^.
instead of n, so as to avoid
31*
484 STATISTICAL PHYSICS [Part IV
kinetic energy could not be compensated by a reduction in magnetic
energy.
Let there be n electrons which have changed their spin directions.
N
Then there remain   n electrons with spin antiparallel to the field,
N
while +n have spins parallel to the field. The limiting energies are
determined from formula (44.8), where we must put^ n instead
N
of g . Whence we obtain the following expression for the limiting
kinetic energy of both types of electrons :
and the equation for the total limiting energies is
/ 3 \'/. (2 */,,)
~
Since n<^ 5, the binomials can be expanded in a series as follows:
z
IN . \'/3
3JV *
Substituting this in (44.14), we find the number of electrons which
change their spin directions in the magnetic field :
(44.15)
Each of these electrons contributes a term 2 p to the total magnetic
moment of the whole gas, because its moment projection on the magnet
ic field has changed from (J to p. The magnetic polarization (that
is, the magnetic moment of unit volume) turns out equal to
while the magnetic polarizability a, defined as the coefficient of H
on the righthand side of this formula, depends only upon the density
of the electron gas and not its temperature :
Indeed, alkali metals have a paramagnetism independent of tempe
rature. Let it be recalled that in accordance with the results of Sec. 40
[see (40.53)] atomic paramagnetism gives a magnetic polarizability
which is inversely proportional to the temperature. Formula (44.17)
agrees satisfactorily with experiment.
Sec. 44]
FERMI DISTRIBUTION
485
Diamagnetism of electrons. L. D. Landau has shown that the quan
tized motion of electrons in a magnetic field this motion is similar
to their classical motion in a spiral leads, in a weak field, to the appear
ance of a magnetic moment equal to 1/3 of expression (44.16), and of
opposite sign. The nature of this effect is purely quantum : if we regard
the motion of electrons as classical then the additional magnetic
moment becomes identically zero (see Sec. 46, exercise 13).
If $H is of the order of 0, then the polarizability does not depend
monotonically upon the field and exhibits much oscillation as the field
increases. The oscillatory variation of magnetic properties is, in fact,
observed in very many metals.
The potential distribution in an atom. We shall now show how to find
the general form for the electrondensity distribution in atoms via
the notion of a Fermi gas. To a certain approximation, the electrons
in heavy atoms resemble a Fermi gas. However, it must be noted that
each electron occurs in the inhomogeneous electric field formed by the
nucleus and the entire configuration of the remaining electrons.
Let us first of all consider a Fermi gas at absolute zero in a potential
field of the form shown in Fig